OpenVMS Notes: "Alpha to Itanium" Porting
Diary
edit: 2018-04-18
Executive Summary- Booting
has changed. Platforms like PDP, VAX, and Alpha presented a
firmware-based BIOS where you could boot the system with commands like
"b dua0" (boot disk dua0)
- All Itanium platforms replace the firmware-based BIOS with a
software-based EFI (or UEFI) targeted at a management processor
- The rx2600 series machines allow you to enter a ROM-based "RAID
management utility" upon boot (usually hitting the F5 key when prompted)
- The rx2800 series machines make it much more difficult (but not
impossible) to access the ROM-based "RAID management utility" during
boot
comment: I'm not not sure when things
changed, but I am now seeing a RAID Setup Prompt on this machine when I
boot OpenVMS-8.4 patch-kit 1200.
The Name Game
- In 1985, Intel began using the phrase IA-32 (Intel Architecture 32)
to market their 32-bit
x86 processor chips
- In 1998, Intel began releasing
Xeon variations of IA-32
targeted at the commercial server market
- In 2001, Intel began using the phrase IA-64 (Intel Architecture 64)
to market their 64-bit
Itanium processor chips
- in 2003, HP began using the name
HP Integrity
to market servers contained Itanium processors (huh?)
- IMHO, this is one reason why most computer professionals today do
not know the name Itanium
- In 2003, AMD added 64-bit extensions to their 32-bit x86 processors
which they marketed as
AMD64
- In 2004, Intel followed AMD's lead by adding 64-bit extensions to their
32-bit x86 architecture which they marketed as
INTEL64
- Now both product lines are collectively known as the
x86-64 architecture
- Because everyone knew that IA-32 meant x86, many just assume IA-64 means x86-64
even though
this Wikipedia article says they are mistaken:
https://en.wikipedia.org/wiki/IA-64
- Things have gotten so bad that most of the Linux community has
adopted this informal naming convention for their distributions:
| Code |
Meaning |
Notes |
| i586 |
Pentium compatible |
"i" because Intel developed it first |
| i686 |
Pentium Pro compatible |
"i" because Intel developed it first |
| AMD64 |
x86-64 compatible |
"AMD" because AMD released it first |
| I64 |
Itanium |
"I64" because it is unique |
- on this page I will only use the name Itanium
A very brief history of Itanium
- Everyone reading this will already be aware of the terms:
- CISC (complex instruction set computer)
- RISC (Reduced Instruction Set Computer)
- VLIW (very long instruction word)
- In the 1989, HP was developing a successor to VLIW which would
eventually be referred to as EPIC (Explicitly parallel instruction
computing)
- Realizing that this would be too big for one company, HP partnered with
Intel in 1994 to develop an EPIC-based chip which would eventually be known
as
Merced
- Merced was expected to appear in 1997 but was delayed until 2001. The
main reason for the delays are two fold:
- RISC and VLIW chips kept getting faster but Merced was not able to
match them
- unlike RISC-based compiler enhancements, expected EPIC-based
compiler enhancements never occurred
- I've often wondered if Merced was intended as nothing more that a
prototype (much like the 8088 was intended to be a prototype for the 8086)
because only a few thousand were ever made. Merced is the only chip
officially known as Itanium. All the chips released after
2001 are officially known as Itanium2
- Inventing Itanium: How HP Labs Helped Create the Next-Generation Chip
Architecture
http://www.hpl.hp.com/news/2001/apr-jun/itanium.html
Project-01 :: HP rx2800-i2
Introduction (my first Itanium)
- I'm not sure where I am going to go with this web page since moving from
Alpha to Itanium should be a whole lot easier than my
previous move from
VAX to Alpha
which involved code renovation (necessary due to sloppy programming in our
group which happened during a decade of "no code reviews")
- 99% of our source code (~ 200 programs: 90% VMS-BASIC; 10% DEC-C) is
built from DCL scripts so I suspect we will only need to use DECnet to
copy over the source code then auto-generate a master script to invoke the
individual
build-scripts
- 1% of our source code involves special odds-n-ends like gSOAP so would
need to be built separately
- I had previously been involved in buying VAX technology from Digital Equipment Corporation
(DEC) as well as Alpha technology from Compaq Computer
Corporation but this was my first time buying Itanium technology from
Hewlett Packard Corporation (HP/HPE)
- OpenVMS software licensing simplified with each owner. With
HP, licensed software under "an active support agreement" has a 100% trade-in
value so is transferred to the new platform at no charge. Licensed software
not under a current support agreement has a 40% trade-in value. In
either case you must supply software authorization numbers.
- What we purchased:
- rx2800-i2 rack-mounted Itanium server (
http://www8.hp.com/us/en/products/integrity-servers/product-detail.html?oid=5033573 )
- one quad-core 9340 Itanium processor
(1.60GHz/5.0MB) which means it is from the
Tukwila
family
- like Windows, OpenVMS sees eight CPUs due to hyper threading
(when they are enabled)
- rx2800-i4 machines are based upon 9500 series Itaniums which
means they are from the
Poulson
family
- 64 GB of RAM
- one built-in P410i Smart Array (RAID-1)
- two 300 GB drives
- one DVD drive
- Caveat: what follows are not full days; this
was a spare-time project I worked on outside my
normal daily duties
Day 1 (well, just a few hours one
afternoon)
-
Itanium rx2800-i2
(drives are horizontal)
The machine arrived two weeks ago by air from Singapore
- It was racked and powered-up by an HP field technician yesterday (June-18, 2015).
This technician knew nothing about OpenVMS but told us that we could have
ordered a machine with a preinstalled OS. We were also told that we could
put in a request to have another HP technician install OpenVMS for a fee (we
declined).
- No paper manuals or optical media were found in the box. Just a letter saying
that online manuals are available at the HP website. The bad news
for me is that this was around the time that HP was
preparing to split into two companies: HP and HPE. The first thing they did
(pre split) was to begin splitting their website which meant that a lot of
documentation I needed was not available at either www.hp.com or
www.hpe.com
-
Anyone coming from an Alpha background will be in for a shock because newer
Itaniums do not have a traditional console. Almost everything is done
via the serial port associated with the
iLO
(integrated lights out) port
- When I first saw this I recalled my introduction to VAX-8550
which employed a
DEC-Pro350 (a personal computer based upon the PDP-11
which ran a modified version RSX-11) as a
console. Although sometimes a pain, having a smaller computer managing a
larger computer offered unexpected advantages; especially during boot (more on this in a moment)
- As an aside, the PDP-11/44 employed an
Intel 8085 to
implement a serial console which included bootstrap support. Management
Processors have been with us for a very long time
BIOS -> EFI -> UEFI
- Personal Computers since the introduction of the IBM-PC in 1981
employed
BIOS firmware to bring
hardware up to a known state before booting an OS.
- as PCs scaled up to
replace traditional minicomputers, BIOS implementations were leveraged to
meet the challenge. Anyone who has ever worked with x86-based commercial servers are already aware of one of the dangers: removing a drive
(or rebooting with a dead drive) will
cause all the lettered drives after it to move up one position in the alphabet. This
causes a real mess if anything boots by, or is mounted by, drive letter.
- Most people reading this will be surprised to learn that BIOS
on modern machines still has the original limits:
16-bit processor mode; dealing with
memory in 64 KB blocks via segment registers but never accessing
more than 1 MB (the actual code limit is 640k); Ad-hoc code is
sometimes tacked on to do boot-time memory tests above 1 MB
- But most important of all, BIOS is almost always was run by the
host CPU
- In the late
1990s, Intel (working with HP on Itanium) began a project
named
Intel Boot Initiative which published a successor technology in 2002 named
EFI
(extensible firmware interface). This is best thought of as a tiny OS
run on a small Management Processor (or simply MP). This management
processor is then used to manage the host environment (see: VAX-8550
comment above)
- In 2005 Intel transferred control of EFI to the
Unified EFI Forum which
now consists of over a dozen companies. EFI was then rebranded
UEFI
-
To learn more:
-
The Management Processor on the iLO of my rx2800-i2 hosts an x86 CPU
-
Anyone who has ever peeked inside any of the EFI files will see:
- the phrase "This program cannot be run in DOS mode" starting at
byte-79 (if you count from byte one)
- see
https://en.wikipedia.org/wiki/Portable_Executable for more
details
- byte-01 and byte-02: x5A4D (little endian)
- byte-75 and byte-76
(little endian values)
- x014C: EFI Image: IA32
- x0200: EFI Image: IA64 (I think this means Itanium)
- x0EBC: EFI Byte Code (think "Java virtual machine")
Itanium rx2650
(drives
are vertical;
beer is optional)
So what's in your Itanium?
- Previous generation Itaniums employed EFI
(sometimes referred to as Effie)
- EFI systems presented a so-called "legacy mode" which is nothing
more than yellow text on a black background.
- You could switch it into
a graphics-based setup display as seen in the rx2650 photo to the right
but certain changes (like enabling/disabling threading) must be done in
"legacy mode"
comment: if you didn't know any
better you would swear you are looking at a picture of an x86-based
DL-380 or DL-385. Notice that these drives are slotted vertically
- Newer Itaniums employ UEFI
- Although my rx2800-i2 does
support VGA graphics, USB-keyboard and USB-mouse, the interface contains
no graphics and appears to be totally text-based. Furthermore, no
provision was made for a booting OpenVMS to communicate with this
display.
- from UEFI you issue the boot command
- your VGA monitor will simply display the phrase: Booting OpenVMS
8.4
- boot messages, including errors and failures, are now only visible on the iLO port
- The VGA monitor, keyboard, and mouse will only become fully
operational after OpenVMS is fully up and has
started an
X-Windows/Motif
terminal. It is for this reason that you should divert the output of
script sys$manager:systartup_vms.com to a log file in
folder sys$manager
- comment: I recently worked on a newer PC (Compaq Proliant 500) which had
a graphics-based UEFI that looked somewhat similar to the picture to the
right.
iLO and UEFI
Data Center Survival Kit
- Trying to find a functional VT monitor with a DB9 serial port
these days is next to impossible so I used my laptop along with a
USB-to-DB9 dongle. In my case both serial ports were DTE (male) so I needed an F-F null
modem cable. TeraTerm
worked like a charm.
- Do yourself a favor and build a Data Center Survival Kit
(you do not want to do this during a "Sunday night emergency" or "a national
holiday")
- You will need:
- an old-school briefcase (engineers never use knapsacks)
- one USB-to-DB9 dongle for your laptop (somehow I ended up with two, one of each sex, but only one
dongle is shown)
- the black dongle shown in the photo is better than average
and has three indicator LEDs
(1-Link, 2-Tx, 3-Rx)
- after you finish installing the driver software from the supplied optical
media, store it behind the document divider in
briefcase
- two null-modem adapters: one Male-to-Male; one Female-to-Female
(orange paint in the photo)
- two straight-through gender changers: one Male-to-Male; one Female-to-Female
(yellow paint in the photo)
- one 3m/10ft straight-through serial cable (Male-to-Female)
pictured top-left
- optional:
- If you work on storage
arrays associated with DL380 and DL385 then you will also need
a DB9-to-micro-DB9 serial cable (pictured top-right). One is usually
supplied with each storage array but they tend to "walk away" if you
know what I mean
- Some real VT-320 monitors employ a 25-pin connector on the
monitor and come with a 25-pin to 9-pin cable. Don't allow these
to walk away
Wasted Time (caused by me)
- to be fair, PDF manual "HP Integrity rx2800 i2 Server User Service
Guide" clearly states on that OpenVMS must be installed on Itanium
using a serial console but I was certain the HP
document was outdated
- surfing around (via Google) will take you to sites claiming that a VGA
monitor could be used to do the initial OS install by first reconfiguring the console
via the CONCONFIG command in the UEFI shell. Perhaps those sites were
installing HP-UX, or perhaps that method works for other Itaniums but it did
not work for me.
- Experiments:
- setting "VGA = primary" and "SERIAL = off"
- failed; VMS came up part way but then displayed an error message about
being unable to continue without a serial port
- setting "VGA = primary" and "SERIAL = secondary"
- failed; VMS came part way up then prompted for date+time. I was
unable to enter it into either keyboard
- setting "SERIAL = primary" and "VGA = secondary"
(this was the way the machine was delivered)
Initial Setup
- optional: connect a VGA compatible monitor, USB-keyboard, USB-mouse
- power-up the system without any media in the DVD drive
- if you are watching from a serial terminal connected to the iLO port,
you will see your first message within a minute
- additional messages will be seen within 2 to 5 minutes depending
upon whether boot-time device diagnostics are disabled or enabled
- you will not see startup messages at the VGA monitor so be patient
- eventually you will see the MP MAIN MENU on both the iLO port
as well as the VGA monitor
- type S to drop into the UEFI shell before
the 7-second timer expires (this delay can be adjusted either way; nothing
wrong with 15 or 20 seconds)
legend for my examples:
<sr> system response (what the system displays)
<ur> user response (what you type or do)
(text) some action is described
<enter> hit the "enter" key
----------------------------------------------------------
<ur> (connect a serial line to the iLO port then hit <enter>)
<sr> MP login:
<ur> Administrator
<sr> MP password:
<ur> ******** (this is found on a pullout tag on the front of your processor chassis)
<sr> Hewlett-Packard Integrated Lights-Out 3 for Integrity
(C) Copyright 1999-2013 Hewlett-Packard Development Company, L.P.
MP Host Name: kawc0x
iLO MP Firmware Revision 01.55.02
MP MAIN MENU:
CO: Console
VFP: Virtual Front Panel
CM: Command Menu
CL: Console Log
SL: Show Event Logs
HE: Main Help Menu
X: Exit Connection
[kawc0x]</> hpiLO->
<ur> co
<sr> (iLo system connects to system Itanium console)
[Use Ctrl-B or ESC-( to return to MP main menu.]
- - - - - - - - - - Prior Console Output - - - - - - - - - -
1,0,0,0 5400006301E10000 0000000000000000 EVN_BOOT_START
***********************************************************
* ROM Version : 01.95
* ROM Date : Fri Feb 01 03:54:28 PST 2013
***********************************************************
1,0,0,0 3400083701E10000 000000000002000C EVN_BOOT_CELL_JOINED_PD
1,0,0,0 340000B101E10000 0000003C0205000C EVN_MEM_DISCOVERY
1,0,0,0 1400002601E10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,1,0 1400002605E10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,2,0 1400002609E10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,3,0 140000260DE10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,1,1 1400002607E10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,2,1 140000260BE10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,3,1 140000260FE10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,0,1 1400002603E10000 000000000006000C EVN_BOOT_CPU_LATE_TEST_START
1,0,0,0 5400020701E10000 000000000011000C EVN_EFI_START
Press Ctrl-C now to bypass loading option ROM UEFI drivers.
1,0,0,0 3400008101E10000 000000000007000C EVN_IO_DISCOVERY_START
1,0,0,0 5400020B01E10000 0000000000000006 EVN_EFI_LAUNCH_BOOT_MANAGER
(C) Copyright 1996-2010 Hewlett-Packard Development Company, L.P.
Note, menu interfaces might only display on the primary console device.
The current primary console device is:
Serial PcieRoot(0x30304352)/Pci(0x1C,0x5)/Pci(0x0,0x5)
The primary console can be changed via the 'conconfig' UEFI shell command.
Press: ENTER - Start boot entry execution
B / b - Launch Boot Manager (menu interface)
D / d - Launch Device Manager (menu interface)
M / m - Launch Boot Maintenance Manager (menu interface)
S / s - Launch UEFI Shell (command line interface)
I / i - Launch iLO Setup Tool (command line interface)
*** User input can now be provided ***
Automatic boot entry execution will start in 7 second(s).
<ur> s
<sr> Searching for devices.
HP Smart Array P410i Controller (version 5.78) 1 Logical Drive
Currently the controller is in RAID mode
Launching UEFI Shell.
UEFI Shell version 2.10 [2.0]
Current running mode 1.1.2
Device mapping table
fs0 :Removable HardDisk - Alias hd9a0b blk0
PcieRoot(0x30304352)/Pci(0x1,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)/HD(1459B1241-18EB-11E5-9F2B-AA000400FEFF)
fs1 :Removable HardDisk - Alias hd9a0d blk1
PcieRoot(0x30304352)/Pci(0x1,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)/HD(3459B1240-18EB-11E5-9F2C-AA000400FEFF)
blk0 :Removable HardDisk - Alias hd9a0b fs0
PcieRoot(0x30304352)/Pci(0x1,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)/HD(1459B1241-18EB-11E5-9F2B-AA000400FEFF)
blk1 :Removable HardDisk - Alias hd9a0d fs1
PcieRoot(0x30304352)/Pci(0x1,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)/HD(3459B1240-18EB-11E5-9F2C-AA000400FEFF)
blk2 :BlockDevice - Alias (null)
PcieRoot(0x30304352)/Pci(0x1F,0x2)/Sata(0x0,0x0,0x0)
blk3 :Removable HardDisk - Alias (null)
PcieRoot(0x30304352)/Pci(0x1,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)/HD(4459B1240-18EB-11E5-9F2B-AA000400FEFF)
blk4 :Removable BlockDevice - Alias (null)
PcieRoot(0x30304352)/Pci(0x1,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)
Press ESC in 1 seconds to skip startup.nsh, any other key to continue.
Shell>
-----------------------------------------------------------------------------------------------------------
Caveat: this listing actually shows two initialized volumes (file systems are prefixed with "fs")
and five block structured devices (prefixed with "blk"). On a brand new system (with no optical
media present in the DVD drive) you will likely see one or more "blk" entries and zero "fs" entries.
Notice how fs0 (red) points to blk0 (red) and vice versa.
Also notice the alias names in green and purple.
Booting the OpenVMS ISO DVD from UEFI
<sr> Shell>
<ur> map
<sr> (system will display known devices; initialized media will appear with fs# entries)
<ur> (insert VMS media into the DVD drive)
map -r
<sr> ("-r" forces the map command to rescan all the hardware)
(the system will display known devices; in my case "fs0:" now refers to the DVD drive)
<ur> fs0:
<sr> fs0:>\
<ur> ls
<sr> (should get a directory listing)
<ur> cd efi\boot
<sr> fs0:\efi\boot>
<ur> bootia64
<sr> (OpenVMS boots then displays the familiar 8-line installation menu)
<ur> (use the menu to install OpenVMS on the desired drive)
Once OpenVMS is installed, eject the DVD then reboot
The system will reboot to UEFI but not OpenVMS (what?)
<sr> Shell>
<ur> map
<sr> (system will display known devices; in my case fs0: now refers to the OpenVMS system drive)
<ur> fs0:
<sr> fs0:>\
<ur> bcfg boot add 1 \EFI\VMS\VMS_LOADER.EFI "OpenVMS 8.4"
<sr> fs0:>\
<ur> exit
<sr> system will drop into the UEFI menu
<ur> B
<sr> system will display boot menu
<ur> choose the first entry (if you entered "add 1" above)
<sr> system will boot OpenVMS
------------------------------------------------------------------------------------------------------
Question: So what just happened?
Answer : The boot configuration command (BCFG) looked at my currently selected device (fs0) then used
associated drive information (1459B1241-18EB-11E5-9F2B-AA000400FEFF in my case) to make an
entry into position #1 of the boot configuration table. Previous entries were pushed down by
one position. Coles Notes Summary: we now boot by GUID rather than drive letter
Day 2 (Installing OpenVMS)
- On page 37 of the 2013 manual HP Integrity rx2800 i2 Server User
Service Guide we read:
For OpenVMS installation, see the VMS84I_UPDATE-V0500.RELEASE_NOTES
document
available in the HP Support Center here: http://www.hp.com/go/hpsc
which I correctly assumed meant we need to install VMS84I_UPDATE-V0500
right now
- I incorrectly thought the simpler route would be to register/load the
OpenVMS License into the license manager then enable DECnet so I could employ the network to copy the patch from another system.
This was a
big mistake. Shortly after I loaded the license I encountered a stack dump.
Every reboot after that ended in a stack dump
- At this point I had two choices:
- reboot the DVD; drop into DCL; mount the disk; connect to the
license manager on that disk to unload the license
- reboot the DVD; reinstall OpenVMS
(I chose this option because, once
started, I could walk away to take care of some other pressing matters
associated with my daily duties)
- Once the system was back up, I used a PC to burn the patch onto a CD-ROM
then used SneakerNet to
get it to the Itanium. I copied the self-extracting patch from the optical
drive then used PCSI to install it. Once rebooted, I registered/loaded the
OpenVMS license; everything worked like a charm
- I was previously worried about the RTC (real time clock)
because every boot of OpenVMS prompted me to input the DATE + TIME at the
serial port console. This stopped as soon as I installed VMS84I_UPDATE-V1000 which happened to be the latest available kit
- As stated previously, you won't see any console messages from script
SYS$MANAGER:SYSTARTUP_VMS.COM on the VGA monitor so it would be wise to
modify the startup script then divert SYS$OUTPUT to file
SYS$MANAGER:SYSTARTUP_VMS.LOG (remember to close the channel before the
script exits)
Day 3
- installing "DECnet phase IV", "TCPIP for OpenVMS", and "Apache" went without a
hitch
- DECnet makes coping files to-from the old AlphaServer-DS20e child's play
Day 4
- installing DEC-BASIC, DEC-C++, and FMS (Forms Management System) went
without a hitch
- we require FMS because we still support a small number of
green-screen apps
- we are still waiting for a DEC-C license which was overlooked during
HP's license trade-in program
Day 5 (playing with RAID)
- When I did the initial OpenVMS install a week ago I noticed that I
was writing to both my disks. This machine was delivered with an integrated
(built into the motherboard)
HP Smart Array P410i controller and someone had already
configured both drives as part of a single RAID-1 logical unit (aka disk mirror). Okay
so redundancy isn't a bad thing but I needed a second logical volume for
files associated with our business (RMS-ISAM databases, MariaDB databases,
Apache, photographs, spreadsheets, etc.) but how do I
manage the RAID array? More importantly, I just purchased two additional
drives and I want to install/configure them without calling HP/HPE
- Page 145 of the manual HP Integrity rx2800 i2 Server User
Service Guide begins by describing utilities not found anywhere on
my system. Many web sites will tell you that a kit containing file
saupdate.efi can be easily located and downloaded
from the HP web site but this was not the case with me
- did I mention that this was the month HP began extensive changes to
their web site as part of their plans to split the company into two?
Much stuff at www.hp.com was unavailable
during the move to www.hpe.com
- So as of Friday
(2015-06-26), any attempts to search for the phrase "P410i" will yield a lot
of information but non of it will be of any use to the OpenVMS community.
For example:
surf over to : http://www.hpe.com (this whole paragraph was updated on 2016-02-11)
click : SUPPORT
click : HPE Servers, Storage and Networking
click : Download Drivers
type : P410i
click : GO
view results : notice lots of ProLiant BL + DL products are listed
click on : HP Smart Array P410i Controller
click : Get drivers, software & firmware
search drop-down: look for "OpenVMS" or "OVMS" (but you will not find either)
select drop-down: "OS Independent"
result : ~100 firmware files but nothing of use to OpenVMS or UEFI
: You will see a lot of Linux stuff here (RPM files, etc.)
surf over to : http://www.hpe.com (this whole paragraph was updated on 2016-02-11)
click : SUPPORT
click : HPE Servers, Storage and Networking
click : Download Drivers
type : HP Integrity rx2800 i2
click : GO
click : Get drivers, software & firmware
click on : HP Integrity rx2800 i2 Rack Optimized Server
select drop-down: "OpenVMS v8.4"
result : 9 firmware files including these 2 in the "Firmware Controller Storage" section:
** CRITICAL ** SAS IO Firmware Update Package (EFI) - HP Integrity rx2800 i2 Servers
which contains this file: rx2800i2IOSASFW578_EFI.zip
** CRI2TICAL ** SAS IO Firmware Update Package for OVMS - HP Integrity rx2800 i2 Server
which contains this file: HP-I64VMS-IOCFM_RX2800I2-V0578.ZIPEXE
(self extracting archive for OpenVMS Itanium)
Day 6 (adding two more drives)
- I installed an OpenVMS patch to make new firmware available so I could
play with the RAID
- running self-extracting archive:
HP-I64VMS-IOCFM_RX2800I2-V0578.ZIPEXE
- yields these two files:
- HP-I64VMS-IOCFM_RX2800I2-V0578--1.PCSI$COMPRESSED;1
- HP-I64VMS-IOCFM_RX2800I2-V0578--1.PCSI$COMPRESSED_HPC;1
- which you install from DCL via the PRODUCT command.
- At this
time EFI files (including: SAUPDATE.EFI) will be copied into a root directory of your boot disk
- you will
be able to see them from the UEFI shell accessible from the iLO port
- Shutdown OpenVMS to play with the RAID
<sr> Shell>
<ur> fs0:
<sr> fs0:>\
<ur> cd efi\tools (look for SAUPDATE.EFI which may be in a different place on your system)
<sr> fs0:\efi\tools>
<ur> saupdate get_mode p410i (or: "saupdate get_mode all")
<sr> The controller at 0:1:0:0 is in RAID mode
fs0:\efi\tools>
<ur> drivers -b
<sr> T D
D Y C I
R P F A
V VERSION E G G #D #C DRIVER NAME IMAGE NAME
== ======== = = = == == =================================== ===============
16 0000000A B - - 1 56 PCI Bus Driver PciBusDxe
1A 00000014 B - - 1 1 Serial 16550 Driver Serial
68 00000001 D - - 1 - ICH Serial ATA Controller Initializ SataControllerD
6E 0000000A B - - 1 1 AHCI Bus Driver AhciDxe
6F 0000000A D - - 3 - Platform Console Management Driver ConPlatformDxe
70 0000000A D - - 2 - Platform Console Management Driver ConPlatformDxe
71 0000000A B - - 3 3 Console Splitter Driver ConSplitterDxe
72 0000000A B - - 2 2 Console Splitter Driver ConSplitterDxe
73 0000000A ? - - - - Console Splitter Driver ConSplitterDxe
74 0000000A B - - 2 2 Console Splitter Driver ConSplitterDxe
75 0000000A ? - - - - Console Splitter Driver ConSplitterDxe
78 0000000A ? - - - - UGA Console Driver GraphicsConsole
79 0000000A B - - 1 1 Serial Terminal Driver TerminalDxe
7A 0000000A D - - 1 - VGA Class Driver VgaClassDxe
7B 00000000 D - - 1 - PCI VGA Mini Port Driver VgaMiniPort
7C 0000000A D - - 5 - Generic Disk I/O Driver DiskIoDxe
7F 0000000A D - - 2 - FAT File System Driver Fat.efi
81 0000000A B - - 1 1 SCSI Bus Driver ScsiBus
82 0000000A ? - - - - Scsi Disk Driver ScsiDisk
Press ENTER to continue, 'q' to exit:
<ur> <enter>
<sr> 83 00000030 ? - - - - Scsi Tape Driver ScsiTape
84 0000000A B - - 1 3 Partition Driver(MBR/GPT/El Torito) PartitionDxe
85 0000000A B - - 7 4 Usb Bus Driver UsbBusDxe
86 0000000A D - - 2 - Usb Keyboard Driver UsbKbDxe
87 00000011 ? - - - - Usb Mass Storage Driver UsbMassStorageD
88 0000000A D - - 2 - Usb Mouse Driver UsbMouseDxe
89 00000020 D - - 5 - Usb Uhci Driver UhciDxe
8A 00000030 D - - 2 - Usb Ehci Driver EhciDxe
8C 0000000A D - - 4 - Simple Network Protocol Driver SnpDxe
8D 0000000A B - - 4 12 MNP Network Service Driver MnpDxe
8E 0000000A B - - 4 4 ARP Network Service Driver ArpDxe
8F 0000000A B - - 4 32 IP4 Network Service Driver Ip4Dxe
90 0000000A D - - 4 - IP4 CONFIG Network Service Driver Ip4ConfigDxe
91 0000000A B - - 24 20 UDP Network Service Driver Udp4Dxe
92 0000000A B - - 4 4 DHCP Protocol Driver Dhcp4Dxe
93 0000000A B - - 8 4 MTFTP4 Network Service Mtftp4Dxe
94 0000000A D - - 24 - UEFI PXE Base Code Driver UefiPxeBcDxe
95 0000000A D - - 4 - Tcp Network Service Driver Tcp4Dxe
96 0000000A ? - - - - iSCSI Driver IScsiDxe
9C 00000350 B X X 1 2 Smart Array SAS Driver v3.50 MemoryMapped(0x
BF 04030100 B X X 1 1 Intel(R) PRO/1000 4.3.01 PCI-E MemoryMapped(0x
C2 04030100 B X X 1 1 Intel(R) PRO/1000 4.3.01 PCI-E MemoryMapped(0x
C6 04030100 B X X 1 1 Intel(R) PRO/1000 4.3.01 PCI-E MemoryMapped(0x
C9 04030100 B X X 1 1 Intel(R) PRO/1000 4.3.01 PCI-E MemoryMapped(0x
Press ENTER to continue, 'q' to exit:
<ur> q
<ur> devices -b
<sr> C T D
T Y C I
R P F A
L E G G #P #D #C Device Name
-- - - - -- -- -- -------------------------------------
5F R - - - 1 56 PcieRoot(0x30304352)
76 D - - 5 - -<UNKNOWN>
77 D - - 2 - -<UNKNOWN>
98 D - - 1 - - PcieRoot(0x30304352)/Pci(0x0,0x0)
99 D - - 1 - - PcieRoot(0x30304352)/Pci(0x1,0x0)
9A B X X 1 2 2 Smart Array P410i Controller ---------+
9D D - - 1 - - PcieRoot(0x30304352)/Pci(0x3,0x0) |
(...snip...) |
<ur> drvcfg |
<sr> Configurable Components |
Drv[BF] Ctrl[BD] Lang[eng] |
Drv[C2] Ctrl[C0] Lang[eng] |
Drv[C6] Ctrl[C4] Lang[eng] |
Drv[C9] Ctrl[C7] Lang[eng] |
Drv[9C] Ctrl[9A] Lang[eng] <------------------------+
<ur> drvcfg -s 9C 9A <---------------------------------------+
<sr> Set Configuration Options
Drv[9C] Ctrl[9A] Lang[eng]
Option Rom Configuration for Arrays (ORCA), version 3.50
Copyright 2009 Hewlett-Packard Development Company, L.P.
Controller: HP Smart Array P410i
Create Logical Drive
View Logical Drive
Delete Logical Drive
- Next, I physically inserted two new drives into the next two spare slots then used the ORCA utility
to include them as part of a RAID-1 set. The whole procedure took less than
one minute
Day 7 (Simplistic Benchmarks)
- Playing with the VMS-BASIC compiler. I compiled the following demo
program on both Alpha and Itanium making sure to get the loop limit from an
external prompt in order to foil certain kinds of automatic compiler optimizations. Also, both
were compiled with command-line switch "/optim=level=0"
1000 !============================================================
! title : neil_100.bas
! purpose: compare executable sizes between Alpha and Itanium
!============================================================
option type=explicit
set no prompt
declare string junk$, long i%, j%
when error in
print "number of loops? (default=10) ";
input junk$
i% = integer(junk$)
use
i% = 0
end when
!
select i% !
case 1 to 100 !
case else ! default
i% = 10 ! default
end select !
for j% = 1 to i% !
print "this is test line:";j% !
next j% !
end !
- And here are the resultant file sizes
| File Name | Alpha | Itanium |
|---|
| NEIL_100.BAS | 1KB | 1KB |
| NEIL_100.EXE | 4KB | 10KB |
| NEIL_100.LIS | 20KB | 34KB |
| NEIL_100.OBJ | 6KB | 6KB |
- For the example program shown above, the Itanium executable is 2.5 times larger. What
follows is data extracted from the listing files
PROGRAM SECTIONS (Alpha)
Name Bytes Attributes
0 $READ_ONLY$ 296 PIC CON REL LCL SHR NOEXE RD NOWRT OCTA
1 $CODE$ 852 PIC CON REL LCL SHR EXE NORD NOWRT OCTA
2 $LINK$ 344 NOPIC CON REL LCL NOSHR NOEXE RD NOWRT OCTA
3 $LITERAL$ 54 PIC CON REL LCL SHR NOEXE RD NOWRT OCTA
4 $ARRAY 0 NOPIC CON REL LCL NOSHR NOEXE RD WRT OCTA
5 $DESC 0 NOPIC CON REL LCL NOSHR NOEXE RD WRT OCTA
PROGRAM SECTIONS (Itanium)
Name Bytes Attributes
0 $READ_ONLY$ 296 PIC CON REL LCL SHR NOEXE RD NOWRT OCTA
1 $CODE$ 1584 PIC CON REL LCL SHR EXE NORD NOWRT OCTA
2 $LINK$ NOPIC CON REL LCL NOSHR NOEXE RD NOWRT OCTA
3 $LITERAL$ 54 PIC CON REL LCL SHR NOEXE RD NOWRT OCTA
4 $ARRAY 0 NOPIC CON REL LCL NOSHR NOEXE RD WRT OCTA
5 $DESC 0 NOPIC CON REL LCL NOSHR NOEXE RD WRT OCTA
- Comparing the listing files:
- the total of all program sections is 1.25 times
larger on Itanium
- the total of the CODE and LINK sections is 1.32
times larger on Itanium
- Interpretation/Extrapolation (2015-07-xx):
- The cores in our rx2800-i2 are running at 1.6 GHz:
- The CPUs in our production AlphaServer DS20e are running at 833 MHz
(our development machine runs at 633 MHz)
- This means the Itanium clock is 1.92 times faster
- Running 1.92 times faster on code 1.25 times larger means that
we should expect an unoptimized BASIC program to run 1.54 times faster
- Running 1.92 times faster on code 1.32 times larger means
that we should expect an unoptimized BASIC program to run 1.46 times faster
- with twice as many cores (four times if you include hyper
threading) I suspect this
new box should, at the very least, be able to do six (4 x
1.50) times more work in
parallel. It is surprising that these same numbers also reflect the all-in
price of a new Itanium being four to six times lower than a new (in 2002) AlphaServer
- Caveat: these overly simplistic comparisons ignore other important
factors like: memory
technology, memory size, cache architecture, cache size, disk drive
transfer rates, maximum NIC speeds, etc.
- Real World Update #1
(2015-07-11): compiling a very large (60k lines) BASIC program on this
Itanium is 2.1 times faster than our production AlphaServer.
- could the faster speed be due to "16 times more more memory"
or "better memory technology" (DDR2 which has double the transfer rate
of DDR)
- Real World Update #2 (2016-02-14): something
wonderful happened after we installed the SAS Cache Memory Module
(see Day20 below). Compiling the large 60k line program
now only requires 35 seconds on this Itanium making it 10.3 times faster
than Alpha. Since this is a "single
process" observation (because the compiler can only engage one CPU core),
having twice as many cores means this
platform should be able to do 20.6 times more work at the
same time. Perhaps even a bit more if you recall that these CPUs are
hyper threaded (OpenVMS sees 8 CPUs).
| Label | Machine | CPU clock | CPUs | Memory
Type |
Memory
Size |
Raid | 60k Line
Compile Time | Single Task Speed Up |
Total
Speedup? |
|---|
| DVLP | AlphaServer-DS20e | 633 MHz | 2 | DDR | 2
GB | Mylex 960 | 361 seconds | n/a | |
| PROD1 | AlphaServer-DS20e | 833 MHz | 2 |
DDR | 4
GB | RA-3000 | 241 seconds | n/a | |
| PROD2 | Itanium rx2800-i2 | 1.6 GHz | one quad
core
(HT=8) | DDR2 | 64
GB | P410i | 35 seconds | 10.3 times faster than
DVLP
6.9 times faster than PROD1 | 20.6
13.8 |
Day 8
- Today I was provided a new LAN circuit to be used as a permanent iLO LAN
connection (you need to use the iLO
serial port to do the configuration)
- This required a reset/reboot
of the iLO (Management Processor) and this was all done with OpenVMS running (very cool).
- By default, connecting to the iLO is done over TCP port 2023 via SSH on
port 22 (note the port number; this is not a typo)
- TeraTerm worked like a charm
Day 9 (Began porting software)
- Today I created directories on the Itanium to replicate our development
environment on the Alpha (we do not have any commercial source code versioning tools
or source code repositories)
[dvlp] - files of the form: yada.bas (VMS-BASIC sources)
yada.com (DCL build scripts)
yada.c (DEC-C sources)
[dvlp.cxx} - files of the form: yada.cxx (DEC-C++ sources)
[dvlp.fil] - files of the form: yada.rec (record maps)
yada.opn (file opens)
[dvlp.fdl] - files of the form: yada.fdl (RMS fdl files used in ISAM tuning)
[dvlp.flb] - files of the form: yada.flb (FMS form libraries)
yada.fms (FMS forms)
[dvlp.fun] - files of the form: yada.fun (VMS-BASIC standalone functions)
yada.sub (VMS-BASIC standalone sub programs)
[dvlp.inc] - files of the form: yada.inc (VMS-BASIC standalone includes)
[dvlp.mar] - files of the form: yada.mar (macro assembler stuff)
[dvlp.sql] - files of the form: yada.sql (SQL files to perform MariaDB maintenance)
yada.com (DCL scripts to perform MariaDB maintenance)
[src] - production copies of everything found in/under [dvlp]
[data] - files of the form: yada.dat (RMS/ISAM production data files)
- Next, I copied over all our source code from the Alpha then prepared
to run the build scripts
- Some very old (not yet modernized) BASIC programs reference VMS run-time
libraries (STARLET.OLB) but the sources do not pick up declarations from BASIC$STARLET.TLB.
These programs pickup their external symbols declarations via an include
[.inc]vms_external.inc but expect to have symbols resolved
at link time where we associate them with "[.mar]symbols.obj"
which is created from DCL by invoking the macro assembler like so:
$mac [.mar]symbols.mar
Even though we are now on a different architecture CPU, this command worked without modification.
comment: I have got to find the time to get rid
of this old crud
- Many of our smaller programs only contain a few external functions found
in [.fun] which are compiled every time we rebuild.
Some larger programs (which are continually being rebuilt) expect to find
the object files associated with precompiled functions inside our custom
library ICSIS_FUNCTIONS.OLB (anyone modifying an external function is
responsible for updating the library). We have been using a home-brewed DCL
script to create/maintain this OLB for the past 15 years starting with VAX
then leading to Alpha. On Itanium this script only works once (library creation) but fails when replacing an individual module upon a subsequent
pass. (you will get a weird message about your OLB not being an ELF
library).
- the fix: modify the library maintenance
script replacing "libr/cre ICSIS_FUNCTIONS.OLB" with "libr/cre/obj
ICSIS_FUNCTIONS.OLB"
- I invoked a few scripts which rebuild all 200+ programs. There were
~25 failures all due to poor housekeeping (eg. they also could not be rebuilt
this way on Alpha)
- the fix: minor modifications to ~25 programs
which then were rebuilt without incident
- I have one program which builds with some very strange linker messages
on Itanium. This area of the code is a place where I employ
sys$cmexec to call a small function which sets/changes a logical
name in table LNM$SYSTEM_TABLE (Why do it this way? It is a supervisor-mode logical
name
which means I must first change over to executive mode; if this stand-alone
program used a CLI then I could just use LIB$SET_LOGICAL; in hindsight the
CLI-based approach might have been less painful).
- the fix:
- modify the called code changing from a BASIC FUNCTION to a BASIC
SUB
- change one declaration in the calling code
Day 10
- due to an administrative FUBAR when we ordered this new platform, DEC-C
was missed. That license arrived today and I was able to build all of our
"C" programs and external functions with no problems whatsoever.
- this excludes some gSOAP apps (haven't installed gSOAP just yet)
- this excludes some MariaDB apps (haven't installed MariaDB just yet)
- this excludes some third-party apps dependent upon MMS (haven't
installed MMS just yet)
Day 11 (Apache woes :: part 1)
- Okay so I didn't see this one coming. Today I downloaded a fresh copy of
CSWS-2.2-1 from the HP website.
- It installed without any errors but moving over the served content was
pointless because the webserver doesn't appear to work
- Our home page contains some jQuery which is employed to convert a
linked-list into a sliding menu (http://www.dynamicdrive.com/dynamicindex1/ddsmoothmenu.htm)
but most other pages at our site also make extensive use of jQuery and AJAX
so if the home page has problems then so will everything else.
- Weird browser behavior
- IE11 seemed to work properly
- Firefox-39 rendered a page but there was no sliding menu, just a
linked list
- Chrome-44 didn't render anything at all but further
examination with the Developer Tool's (hit F12) showed a problem with
incomplete chunked transfers
- Rather than waste your time telling you all the things I did which did
not work, let me say that CSWS-2.2-1 now requires text files to be stored in
stream_lf format
- We were running CSWS-2.2 on the Alpha and it did not require stream_lf format
text files so it looks like we're in a jam for a little while. What to do?
- load CSWS-2.1 then start applying update patches
-OR-
- convert all text files to stream_lf (this includes software
generating new static html-based reports as well as interfaces
responsible for capturing text-file uploads)
Day 12 (Apache patch)
- Good news. I requested a "CSWS-2.2-1 patch" from from HP today via my
OpenVMS support account. HP responded within an hour by placing the patch in
a 24-hour drop box. I installed the patch ( a replacement version of
APACHE$HTTPD_SHR.EXE ) and now everything works properly.
- I downloaded then installed a new version of DDSMOOTH from here:
http://www.dynamicdrive.com/dynamicindex1/ddsmoothmenu.htm
- comment: anyone working with VMS or OpenVMS
is probably aware of FTP-induced corruption were some
files (like .js) might be saved with a record format of "512-fixed"
rather than "variable length". Apparently that was what had happened
with the previous version of DDSMOOTH on Alpha but we didn't know it.
This time I downloaded the files to my PC, zipped them, then FTP'd the
zip to OpenVMS. Unzipping text files on OpenVMS almost always saves them
as stream_lf which is not a problem for editors and/or
Apache.
- Installed Mark Berryman's port of
MariaDB (an alternate fork
of MySQL)
Day 13 (my FTP woes)
I think it is safe to say that most modern systems push messages between them
using SFTP (which is based upon SSH) or SOAP. However, older interfaces connected to other non-OpenVMS
systems may require FTP transfers and this can be a problem. Why?
- almost all out-of-the-can software can not make sense of VMS/OpenVMS
file listings as viewed from FTP (remote systems expect to see a UNIX
display)
- this is especially true of almost all graphical FTP tools
available for PCs (Attachment's Reflection for UNIX and
OpenVMS is one notably expception)
- scripting is virtually identical on all implementations of SFTP
- scripting is implementation specific with FTP (most remote systems expect
to see UNIX)
- the only things that FTP and SFTP
have in common are three letters (this prevents others from using an
SFTP client to connect to our FTP server)
- while FTP supports ASCII-mode transfers, SFTP does not
- FTP scripting will allow you to include a password in the script,
this was removed from SFTP for obvious reasons
Okay so it turns out that we have two separate FTP-related issues which may
force us to a third-party stack (MultiNet)
- Some foreign systems connect to our OpenVMS system via FTP and expect to
see a UNIX-like file listing. These are programmatic transfers written using
Microsoft's C# making use of the FTP API built into
.NET -AND- we have had no luck
getting those programmer's to change their code (they push-pull between
various systems of which we are only one of many)
- both TCPware
and MultiNet
provide logical names (eg.
multinet_ftp_unix_style_by_default is one of many) to
adjust the view as seen by connecting clients. This is done on a
per-account basis by inserting ftp-specific logical names into the
receiving account's LOGIN.COM script.
- TCPIP Services for OpenVMS does
not have anything like this.
- TCPIP Services does employ a system-wide logical name (
TCPIP$FTPD_ UNIX_DISPLAY ) but it does not do what you might think
(it only allows filenames with multiple dots to be displayed that
way)
- We are required to do programmatic semi-intelligent FTP transfers to
other sites so need an FTP API or FTP scripting capabilities. Here is the
pseudo code:
- prepare to push a message file to a remote system
- connect to the remote system via FTP
- look to see if file "YADA.txt" exists
- If NO, then push our message
- If YES, then the remote system has not picked up previous file so don't
overwrite it but try again in 60 seconds
- If still YES, then raise an alarm
- If this interface were rewritten today we would demand to push
uniquely named files of the form "YADA-ccyymmddhhmmss.txt" but would
still need to deal with problems like: "full destination disk", "unable
to connect", and "receiving end not processing their inbound
message files"
(our) Alpha Environment
Our existing AlphaServer DS20e employs the TCPware stack
- many of our inbound FTP accounts employ special ftp-specific logical
names to make our system appear UNIX-like
- TCPware provides an "FTP Library" which is an FTP API
used by high-level languages (BASIC and C in our case) to do outbound programmatic FTP
transfers
- Some outbound FTP transfers still employ the programmatic scripting (referred to as a
"take" file)
(my) Itanium Options
- get a TCPware license for this Itanium
- minus: extra money
- plus: very few (perhaps no) changes required in our programs or
scripts
- minus: TCPware is stuck at IPv4 and will never be upgraded to IPv6
according to the vendor (not currently a problem but could be within a
year or two)
- get a MultiNet license for this Itanium
- minus: extra money
- plus: like TCPware, MultiNet supports special logical
names to modify how FTP clients see our system
- plus: supports both IPv4 and IPv6 (unlike TCPware)
- minus: FTP Library is gone so no FTP API to do FTP from high level languages
(which means that any program doing FTP will need to be rewritten using
simple scripts)
- minus: does not support programmatic scripting (a "/take" switch
exists but works like the "/input" switch found in TCPIP Services)
- any solution developed here would need to be really dumb
- stay with TCPIP Services for OpenVMS
- plus: free with OpenVMS license
- plus: supports both IPv4 and IPv6 (unlike TCPware)
- minus: no logical names (per-account or system-wide) to present UNIX-like displays
to FTP clients
- we would probably need to get a second UNIX or WINDOWS box which
would be designated as designation for FTP clients. This could get
messy
- minus: no support for programmatic FTP scripting
- any solution developed here would need to be really dumb:
- invoke script to connect to remote system to look for
presence of desired file, then exit
- make some decision
- invoke a different script to do the next step
Day 14 (human-caused problems with SFTP)
- Received a trial license from Process Software then downloaded and installed
MultiNet
- As usual, I was interrupted when configuring SSH which led to some
delays in getting SFTP to work properly.
- I believe this very misleading error message
Disconnected; protocol error (Protocol error: packet too
long
would have happened with any stack from any vendor which
is why I have documented the problem
here.
- I think I found a problem with FTP UNIX mode. The MultiNet version of
the process logicals are supposed to work the same way as they do in TCPware but they
don't. So check out this stub from LOGIN.COM script of one of the accounts
needing this
$! logicals for TCPware
$ DEFINE TCPWARE_FTP_UNIX_STYLE_BY_DEFAULT TRUE
$ DEFINE TCPWARE_FTP_DISALLOW_UNIX_STYLE FALSE
$ DEFINE TCPWARE_FTP_UNIX_STYLE_CASE_INSENSITIVE TRUE
$ DEFINE TCPWARE_FTP_UNIX_YEAR_OLD_FILES TRUE
$ DEFINE TCPWARE_FTP_STRIP_VERSION TRUE
$! logicals for MultiNet (our Itanium box)
$ DEFINE MULTINET_FTP_UNIX_STYLE_BY_DEFAULT TRUE
$!~~~ DEFINE MULTINET_FTP_DISALLOW_UNIX_STYLE FALSE (bug - must be undefined)
$ DEFINE MULTINET_FTP_UNIX_STYLE_CASE_INSENSITIVE TRUE
$ DEFINE MULTINET_FTP_UNIX_YEAR_OLD_FILES TRUE
$ DEFINE MULTINET_FTP_STRIP_VERSION TRUE
Day 15 (gSOAP)
- I found some spare time today to install gSOAP; this took no time at all
- Next, I used my scripts to build the client and server apps which were
written in "C"; this took even less time and was totally uneventful (and the
client program needed to link against the OpenSSL libraries)
- comment: by why should I be surprised? Bell
Labs created the "C" language with portability in mind
Day 16 (problem with the Itanium Linker)
- Numerous people (including me) have been crabbing that OpenVMS should be
more UNIX-like. We all got our wish with the Itanium Linker whose origins
begin in UNIX SVR4 as is briefly described here:
Executable and Linkable Format (ELF)
- Our system has several apps which "build properly via the MMK (make) tool on
Alpha" but fails on IA64 with the following link error:
%ILINK-E-INVLDHDR, invalid ELF header; field 'ehdr$q_e_phoff' has invalid
value %X0000000000000048
- My link error was associated with linking against sys$share:libzshr.exe so I located then installed a new version of
ZLIB (v1.28) but this did not help
- Since ZLIB is built using MMK I located then installed a new version of
MMK (v5.1) but this did not help
- At this point browsing for ehdr$q_e_phoff yielded
a lot of hits relating to ELF MAGIC NUMBERS which resulted in me doing "$ana/obj
libzshr.exe" and "$dump/page libzshr.exe" (where I was able
to see "x48" in the correct position)
- This is one of the new ELF parameters you will see
documented in ELFDEF (elfdef.h) along with: ehdr$b_ei_mag0, ehdr$b_ei_mag1,
ehdr$b_ei_mag2, ehdr$b_ei_mag3, ehdr$b_ei_class, ehdr$b_ei_data,
ehdr$b_ei_version, ehdr$b_ei_osabi, ehdr$b_ei_abiversion
- Next I performed the previous "$ana" and "$dump"
commands on file TCPIP$IPC_SHR.EXE which is a working share on my system but
did not see any important differences between this file and libzshr.exe
- Next I began browsing for documents where people might have experienced
problems linking against TCPIP shares on Itanium which led to an ECO with this
caveat:
shares must always appear listed as one-liners after "sys$input/opt"
- The first example below is the equivalent link command generated by our
MMK and does not work
- The second link below contains the suggested modification and does work
without errors or warnings
- Note: it would have been just as legal to insert "sys$input/opt"
just above the last line. However, I have seen different warnings about
not sending more than 900 characters to this compiler "on one equivalent
line" where example-1 already exceeds 600 bytes. The equivalent line
length of example-2 is ~200 bytes which ends just after "sys$input/opt"
$! link-1 does not work on Itanium
$! yields: 1 error and 4 warnings
$ link /exec=qsrv.exe -
[dvlpc.qsrv.objs]qsrv.obj, -
[dvlpc.qsrv.objs]cgi.obj, -
[dvlpc.qsrv.objs]cdescriptor.obj, -
[dvlpc.qsrv.objs]cquery.obj, -
[dvlpc.qsrv.objs]cstatus.obj, -
csmis$dvlp:ICSIS_FUN_LIBRARY.OLB/library, -
mysql055_root:[lib.ia64]libclientlib.olb/library, -
mysql055_root:[lib.ia64]libdbug.olb/library, -
mysql055_root:[lib.ia64]libmysys.olb/library, -
mysql055_root:[lib.ia64]libsql.olb/library, -
mysql055_root:[lib.ia64]libstrings.olb/library, -
mysql055_root:[lib.ia64]libvio.olb/library, -
mysql055_root:[lib.ia64]libyassl.olb/library, -
sys$share:LIBZSHR.EXE/share
$! link-2 does work on Itanium
$! yields: 0 errors and 0 warnings
$ link /exec=qsrv.exe -
[dvlpc.qsrv.objs]qsrv.obj, -
[dvlpc.qsrv.objs]cgi.obj, -
[dvlpc.qsrv.objs]cdescriptor.obj, -
[dvlpc.qsrv.objs]cquery.obj, -
[dvlpc.qsrv.objs]cstatus.obj, -
sys$input/opt
csmis$dvlp:ICSIS_FUN_LIBRARY.OLB/library
mysql055_root:[lib.ia64]libclientlib.olb/library
mysql055_root:[lib.ia64]libdbug.olb/library
mysql055_root:[lib.ia64]libmysys.olb/library
mysql055_root:[lib.ia64]libsql.olb/library
mysql055_root:[lib.ia64]libstrings.olb/library
mysql055_root:[lib.ia64]libvio.olb/library
mysql055_root:[lib.ia64]libyassl.olb/library
sys$share:LIBZSHR.EXE/share
Day 17 (Apache woes :: part 2)
Overview
- For the past three years we have support web-based file uploads to our
OpenVMS-8.4 Alpha platform.
- These files (both text as well as binary) are sent by contractors
and can be as large as 500 MB
- we can go higher but have decided to block anything over 500 MB
because we haven't gotten around to buying more drives
- Anyway, this is all done with a little bit of HTML, CGI and CSWS-2.2
(we employ the HTTP POST method)
New Problem
- We are currently running CSWS-2.2-1 on the new Itanium and I just
learned that we can't upload any file greater than 61 KB
- Thinking this could be a MultiNet problem, I flopped our stack back
to TCPIP Services where the limit jumped to 140 KB
- Our code was written in VMS-BASIC so I quickly banged out a
C-based
test-tool just to convince myself this problem had nothing to do with the
Basic-RTL
- Downgrading to CSWS-2.2 fixed the problem (HP was notified on
2015-10-08)
Curiosity Killed The Cat (satisfaction brought it back)
- I've got an old Alpha junker sitting in the corner with OpenVMS-8.4
installed. Just out of curiosity, I installed the Alpha version of
CSWS-2.2-1 and it worked properly.
- So this problem only affects Itaniums
Quick Fix (2015-11-13)
- HPE just informed me that the problem has been isolated to module
APACHE$APR_SHR.EXE and that the module from 2.2 can be used to make 2.2-1
work properly
Final Fix (2015-11-27)
Day 18 (important miscellaneous stuff)
Third Party Software (we can't live without)
- DELIVER
- MUNPACK (ver 1.5)
- we need this to gracefully extract attachments from MIME-based
emails
- http://www.decuslib.com/DECUS/freewarev50/mpack15/
- supports VAX, Alpha
- modify these
two files then invoke a make tool like mms or mmk (we use
mmk) and a c-complier
- caveat: other sites offer MPACK/MUNPACK versions 1.6a and 1.6b which
do supply Itanium binaries. However, they add new UNIX functionality for
all OSs (like not writing out a file if it already exists). This isn't
an issue on OpenVMS systems due to the ability of Files-11 to include a
file version number. Anyway, if you use 1.6 then you must supply an
additional command line switch to disable the new features.
- ZIP + UNZIP
- For the time being we will be using BACKUP to create nightly
savesets which need to be zipped before sending across a network to our
archival platform in another city. Very large savesets can only be
zipped by an updated version of zip. Your can get ZIP-3.0 and UNZIP-6.0
here: ftp://ftp.info-zip.org/pub/infozip/vms/
but really large savesets should be compressed using gzip which you can
get here: http://antinode.info/dec/sw/gzip.html
(gzip is much faster than either bpzip or bzip2)
Other
- SYSUAF Proxy
- Our webserver makes proxied DECnet connections to our Alpha (no
password is sent across when we do RMS file operations over the DECnet
cable) so you only need to remember how to enable the proxy feature in
SYSUAF then bang them in
Day 19 (supported drives limitations)
Day 20 (drive limitations removed)
- we ordered an SAS Cache Memory Module 512 MB
(578882-001) and Cache Super Capacitor and Cable
(587324-001)
- we installed them will little difficulty (experienced
computer room techs with more than 30
years experience)
- contrary to the manual, the capacitor clips onto the PCI riser
assembly just above the slot holding the SAS Cache Module
- the capacitor cable plugs into a little socket on the SAS Cache
Module but you won't see it until you remove the module from the
anti-static envelope
- you remembered to wear your anti-static strap, right?
- after power-up we saw console messages about firmware seeing the cache
then charging the capacitor (I think it is the equivalent of one 8
Farad capacitor traditional power supplies would first see as a dead
short; hence the need to have software and hardware perform a controlled
charging operation)
- that was not a typo; it is the equivalent of 8,000,000 uF
- running ORCA from the UEFI Shell (see above) allowed us to configure two
new stand-alone drives
- we now have 6 drives
- slots 1 + 2 are joined as one RAID-1 volume set
- slots 3 + 4 are joined as one RAID-1 volume set
- slot 6 is a backup copy of our system disk which will only be used
during emergencies
- slot 5 is a standalone scratch to be used as the destination for
numerous backup operations from DCL
- Our backup strategy has changed. Since this machine has 4 NICs
capable of running 1 Gb/s we grabbed two of them to create:
- a private DECnet network which only runs through our
computer room
- a private TCP/IP network which also runs through our
computer room as well as to an adjacent office
- a Windows-7 PC's in the office is running a server
instance of
FileZilla and is connected to this network
- we purchased ten One Terabyte USB
drives which will be the destination for unzipped savesets
- it is someone's job to change the drive every day,
update a paper log, then courier the driver to another
location across town
- During an emergency recovery someone only needs a PC to FTP
savesets back to the recovery host
- observations:
- now all disk operations are noticeably faster but I can't provide you
with any before-after benchmarks
- if you are using the SAS
drives for primary storage then install this kit even if you do not need
additional storage. The increased performance is worth the price.
- when I first heard the phrase "super capacitor" I assumed it was
just a marketing phrase but I was wrong (click:
https://en.wikipedia.org/wiki/Supercapacitor
). Hopefully this change by HP will put an end to the constant
reordering of replacement 4.8V MEH Batteries found in
their onboard RAID controllers from the mid-2000's
Day 21 (the cutover)
- The cutover was totally uneventful. A 100% success.
- Well, perhaps only 99%. We missed a few rinky-dink items caused by
lapses in our own documentation. This is an area we intend to improve.
- The cut-over was done on a Sunday, and the following Monday was a
national holiday so only a skeleton staff used the system. Thus means
that 99% of the rinky-dink stuff was fixed before before the clients hit
the system on Tuesday
- Most clients only
noticed everything was a lot faster (perceived slow-downs are almost always
related to the corporate Intranet or client PCs)
What a long strange trip this has been...
- Hardware
- this home-grown soup-to-nuts business system started life on a used
32-bit VAX-730 in 1991
- maintenance being the poor kid in town, our system migrated to
computer hardware abandoned by other departments including: VAX-750, uVAX-3500,
uVAX-4300, then finally VAX-6430 (while hardware was free we always had to buy software licenses)
- it was migrated to a used AlphaServer-4100 in 2001
then to a new AlphaServer-DS20e in 2002
- it was migrated to a new Itanium rx2800-i2 in 2016
- Software 1991
- our data was 100% ISAM (RMS)
- our code was 90% VAX-BASIC linked to FMS (forms management system)
with 10% DCL
- we only employed DECnet with a little XCOM
- TCP/IP came in 1992 in
the form of TCPware (there where no TCP/IP products available from DEC
for VAX/VMS in those days)
- ~200 users employed VT-220 terminals to access the system via serial
connections (dialup modems, dedicated modems, limited distance modems, LAT-based terminal
servers)
- Software 2016
- our data is now 50% ISAM (RMS) and 50% Relational (MySQL/MariaDB)
- our code is 60% VMS-BASIC, 10% DEC-C/C++, 10% DCL, 20% JavaScript
(including jQuery, and AngularJS)
- ~1400 users employ three major browsers (Chrome, Firefox, IE11) to access
the system via TCP/IP
- they authenticate against a home-brewed application called the
ProfileDB which allows us to have longer user names than would be
possible with SYSUAF
- green-screen access is restricted to developers and system admins
Summary (so far)
- What a neat platform. Upon first blush this
rx2800 appears
ten times more powerful
than our existing AlphaServer DS20e but at one quarter the cost.
- by cost I mean both HP hardware (rx2800-i2) and licensed HP software (OpenVMS,
DEC-C, DEC-C++, VMS-BASIC, FMS-DVLP)
- production-only platforms will reduce software costs by not
including licensed development tools like compliers
- others might save more by purchasing used hardware from the
third-party market
- OpenVMS boots in 36 seconds from the UFEI shell prompt
(includes starting DECnet IV, TCPIP Services, DECwindows/Motif, Apache,
and MariaDB)
- I previously remember saying: I didn't think there was any argument
(outside of not being in possession of your source code) for not porting
"from VAX to Alpha".
- Now I am forced to say the same
thing about porting "from Alpha to Itanium". In fact, I highly recommend it.
- While I have no doubt that our platform will eventually be migrated again to
OpenVMS on
x86-64 when VSI finishes their OS port,
our 13-year-old Alpha was starting to give us some grief so we just
couldn't wait any longer.
Project-02 :: rx2660
Day 1 (fetch units; initial power up tests)
- I just acquired two rx2660 servers were about to be junked by
another department so I jumped in the van to retrieve them before they made
it to the crusher
- All the drives had been degaussed then crushed for security reasons
so I ordered four new SAS 146 GB drives to begin testing
- After connecting a VGA monitor and USB keyboard, the servers boot to a GUI setup screen which can be flipped to so-called
legacy mode
- GUI mode: white text on a blue background at
the top; blue text on a gray windows below
- Legacy Mode: yellow text on a black
background
- Specs:
- firmware was published in 2008
- two dual-core CPU modules
- 1.65 GHz clock with 9 MB of cache (
Montecito
)
- 16 GB of memory
- two RJ-45 style 10 Base-T/100 Base-T/1000 Base-T system LAN ports
- iLO (integrated lights out module)
- includes a private LAN port
- two hot-swappable power supplies
- What a neat system! It is almost as easy to set up as a new PC (more or less)
- Like VAX before it, this appears to be the end of AlphaServer technology in our shop;
but OpenVMS still lives here!
Day 2 (installing drives; installing OpenVMS)
- New drives arrived today so I slotted two of them.
- After inserting my two I can still only see the DVD from the
EFI Shell.
- You can run "map -r" all day but will not see the new drives until you
run the ORCA (Option Rom Configuration for Arrays) Utility to tell the P400
controller that the drives are, or not, part of a RAID set. The fastest way
to do this is to run "bcfg" (boot configuration).
- Once this was done, I booted the OpenVMS DVD
- Like the rx2800 above, on a VGA monitor you will see ~ 120 dots (on one
and a half lines) at which point you must continue with a serial connection to the
iLO port
- Like the rx2800 above, the real-time-clock will continue to prompt for
the date-time at boot until "you install the load an OpenVMS license" -AND-
"install OpenVMS-8.4 update 500"
- Unlike the rx2800-i2 above, this unit will recognize more than four
drives without the requirement for optional RAM (neat!)
Day 3 (updating firmware and other stuff)
- Our group has an active software support agreement with HP/HPE which
allowed me to download the HPOFM (HP Offline Firmware Management) utility.
This is an ISO image which you burn to a CD-ROM then boot in your Itanium
server. This utility advanced my firmware dates from 2008 to 2013 in ~ 7
minutes. Slick!
- caveat for anyone who is tempted to work on the
CPU:
- each CPU is installed in a ZIF (zero insertion force) socket. This
means you must used the provided wrench to unlock the socket pins before
removing the CPU
- you must also unlock the socket pins before installing a new unit
unit then lock the pins afterward.
- see page 48 of HPE Integrity rx2660 Server User Service
Guide for the details
- well that's all I'm going to post here. This is slick technology and I
wished we could have used it to replace our Alpha Servers 5-years ago.
Day 4 (optional: Installing Linux on Itanium)
- I just downloaded a minimal CD image of Gentoo 20160826 (these are
continually rolling updates so always include
the published date)
- Linux installation details where moved
here:
Installing Linux on Itanium
(Gentoo, SuSE, Debian, etc.)
Common Stuff
Comparing UEFI to EFI
At first blush you might think that the graphics-based setup associated
with EFI systems may be of some use to you but you would be wrong.
- many
things can be seen in GUI mode (like whether "CPU threading" is enabled or
disabled) but can only be changed from text mode (also called legacy
mode).
- this can be "text mode on a VGA monitor" or "text mode on a
character cell terminal over the iLO port"
- there is a boot command associated with EFI but the only way to boot
this UEFI system is via the bcfg (boot configuration) utility.
- don't waste your time with the VGA monitor unless you are doing a
training session; the fastest way to set up any Itanium system is over the iLO
port
Monitoring RAID drives with MSA
During my computer-room walk-through yesterday (2017-11-13), I noticed one dead drive (#6) on KAWC09.
This got me thinking: what was the best way to monitor firmware-managed drives from the OS?
What follows is my first stab hack at this.
##########
Facts:
1) Our rx2660 hosts a P400 RAID controller.
2) Our rx2800-i2 hosts a P410i RAID controller.
3) These controllers can be set up for: JBOD (just a bunch of drives), RAID-0, or RAID-1.
4) If an optional license is entered (DVLP has such a license), the controller can also support RAID-5 and higher.
5) The controllers will be seen as “PK” devices in OpenVMS
##########
Here is my first stab at solving the monitoring problem
KAWC09(DVLP)::Neil> sh dev pk/full
Device PKB0:, device type HP Smart Array, is online, error logging is enabled.
Error count 0 Operations completed 706
Owner process "" Owner UIC [SYSTEM]
Owner process ID 00000000 Dev Prot S:RWPL,O:RWPL,G,W
Reference count 0 Default buffer size 65535
Current preferred CPU Id 7 Fastpath 1
Current Interrupt CPU Id 7
Device PKA0:, device type HP LSISAS106x SAS, is online, error logging is
enabled.
Error count 0 Operations completed 1
Owner process "" Owner UIC [SYSTEM]
Owner process ID 00000000 Dev Prot S:RWPL,O:RWPL,G,W
Reference count 0 Default buffer size 0
Current preferred CPU Id 6 Fastpath 1
Current Interrupt CPU Id 6
SAS Address 5006-0000-0001-C270
KAWC09(DVLP)::Neil> sas :== $sys$system:sas$util.exe
KAWC09(DVLP)::Neil> msa :== $sys$system:msa$util.exe
KAWC09(DVLP)::Neil> msa
MSA> set controller pkb0
MSA> sho disks
SAS device [Disk]
Disk 8: bus: 1, Target id: 0, Port: 1I, Box/Enclosure: 0, Bay: 8, size 136.73 [146.82]GB
Disk 8, # 0, size 286677120 blocks, (136.70 [146.78] GB), Unit 2.
Disk 8, # 1, size 6832 blocks, (3.34 [3.50] MB), Unused.
SAS device [Disk]
Disk 7: bus: 1, Target id: 1, Port: 1I, Box/Enclosure: 0, Bay: 7, size 136.73 [146.82]GB
Disk 7, # 0, size 286683904 blocks, (136.70 [146.78] GB), Unit 1.
Disk 7, # 1, size 48 blocks, (0.02 [0.02] MB), Unused.
SAS device [Disk]
Disk 5: bus: 1, Target id: 3, Port: 1I, Box/Enclosure: 0, Bay: 5, size 136.73 [146.82]GB
Disk 5, # 0, size 286683904 blocks, (136.70 [146.78] GB), Unit 1.
Disk 5, # 1, size 48 blocks, (0.02 [0.02] MB), Unused.
SAS device [Disk]
Disk 4: bus: 1, Target id: 4, Port: 2I, Box/Enclosure: 1, Bay: 4, size 136.73 [146.82]GB
Disk 4, # 0, size 286683904 blocks, (136.70 [146.78] GB), Unit 1.
Disk 4, # 1, size 48 blocks, (0.02 [0.02] MB), Unused.
SAS device [Disk]
Disk 3: bus: 1, Target id: 5, Port: 2I, Box/Enclosure: 1, Bay: 3, size 136.73 [146.82]GB
Disk 3, # 0, size 286683904 blocks, (136.70 [146.78] GB), Unit 1.
Disk 3, # 1, size 48 blocks, (0.02 [0.02] MB), Unused.
SAS device [Disk]
Disk 2: bus: 1, Target id: 6, Port: 2I, Box/Enclosure: 1, Bay: 2, size 136.73 [146.82]GB
Disk 2, # 0, size 286683904 blocks, (136.70 [146.78] GB), Unit 1.
Disk 2, # 1, size 48 blocks, (0.02 [0.02] MB), Unused.
SAS device [Disk]
Disk 1: bus: 1, Target id: 7, Port: 2I, Box/Enclosure: 1, Bay: 1, size 136.73 [146.82]GB
Disk 1, # 0, size 286677120 blocks, (136.70 [146.78] GB), Unit 0.
Disk 1, # 1, size 6832 blocks, (3.34 [3.50] MB), Unused.
MSA> help
Information available:
ACCEPT ADD DELETE EXIT FLASH HELP HPSAS
LOCATE MSA Overview READ RESET SCAN SET
SHOW SmartArray START
Topic?
MSA> exit
KAWC09(DVLP)::Neil>
Notes:
- This RAID array defines 8-drives as three logical units
- Notice that drive-6 (which is dead) is missing from this list
- use command MSA to inspect the physical drives
- use command SAS to inspect the logical units
Used MSA$UTIL to replace a drive on a running system
My production OpenVMS system runs on an rx2800-i2 where the P410 controller
is set up like this:
disk slot-1 +---- RAID-1 (system disk)
disk slot-2 +
disk slot-3 --+-- RAID-1 (data disk)
disk slot-4 --+
disk slot-5 empty
disk slot-6 empty
disk slot-7 ----- RAID-0 (scratch data)
disk slot-8 ----- RAID-0 (emergency system disk)
I rebooted the system on Sunday 2018-02-18 (annual maintenance: installed consolidated
OpenVMS patch kit 1200; updated to MultiNet-5.5). During reboot I noticed
several messages from the
S.M.A.R.T. subsystem indicating an imminent failure of the drive in
slot #7 could be expected soon. I ordered a replacement drive which came in two
days later. I swapped out the drive but nothing happened. At this point, I fired
up MSA$UTIL and noticed some identification problems with Unit-2. At this point I did the
following:- used "delete" to erase the definition of Unit-2
- used "set" to create a new definition of Unit-2
At this point
everything worked properly from OpenVMS (was able to mount the disk). In hind
sight, perhaps I should have experimented with the "accept" command.
Backup and Restoration suggestions
Caveats: the following advice is targeted at all computer systems this side of
Y2K: Unix, Linux,
OpenVMS and Windows.>No matter how you interpret what follows, you must ALWAYS do whole disk
backups of all your drives. Why? You never know when your computer room
will be damaged by fire, water, or anything else which will require you to
recover your system at a new location. Make sure you are do daily
backups (at the very minimum) and deliver the external media to an off-site
location. This means buying enough media to rotate through 30-days at the very
least.
comment: I know of one company where their Windows systems were
infected on day x then the malware was activated 3-months later. This means that
the backups were also infected.
Advice for OpenVMS on Itanium
Facts:
- Most, if not all, system disks are now partitioned. This is especially
true of OpenVMS on Itanium.
- On a two-partition system, one partition will be dedicated to EFI (which is
FAT-formatted) while the other will be dedicated to OpenVMS and is usually
ODS-5 formatted
- Poking around DCL on OpenVMS will show you that no provision has been
made for disk partitioning.
- In order to partition a new system disk you will need to boot the
DVD then build a system disk from that environment.
- From that point you would then restore your OpenVMS backup from tape (or
whatever) only to discover that your whole disk backups were of the OpenVMS
partition only and not the EFI partition (oops)
Suggestions
- Ensure you never need to do an OpenVMS restore (except during an
emergency resurrection) by ensuring that your system disk is fully backed up
via the RAID.
External Links
Back
to
OpenVMS
Back
to
Home
Neil Rieck
Kitchener - Waterloo - Cambridge, Ontario, Canada.