Difference between revisions of "Altix 350"

From Nekochan
Jump to: navigation, search
m (Operating System Support: - Adding CentOS 5.x info.)
m (Troubleshooting and Repair: - PSU failure issue)
Line 82: Line 82:
 
=== Replacement Parts ===
 
=== Replacement Parts ===
  
The Altix 350 uses a common Delta Electronics DPS-500EB modular power supply. While the DPS-500EB E variant is most commonly reported by 350 owners, others have successfully substituted DPS-500EB A units.
+
The Altix 350 uses a common Delta Electronics DPS-500EB modular power supply. While the DPS-500EB E variant is most commonly reported by 350 owners, others have successfully substituted DPS-500EB A units. However there is a known failure mode for earlier versions of these PSUs when run 24/7 for extended periods. See [http://forums.nekochan.net/viewtopic.php?f=3&t=16725720 this thread] for details and a fix.
  
 
=== Mixed PROM Versions ===
 
=== Mixed PROM Versions ===

Revision as of 19:54, 6 July 2013

Industry Leading...
A single Altix 350 node
Half rack of Altix 350 nodes
A full rack of Altix 350 nodes and two routers

This page captures information specific to the Altix 350 model. Please see the very detailed Altix series overview page.

SGI Introduced the Altix 350 model at LinuxWorld in New York on January 20th, 2004. This model was a midrange complement to the Altix 3000 line introduced the previous year, and could be configured with as few as one Itanium 2 CPU in a single 2U module or "brick," to as many as 32 CPUs across 16 modules in a full 39U rack. Meanwhile the 350 provides the same single system image and software features found in the larger configurations, and runs the same operating systems and applications. The model was also offered as a cluster node using Ethernet and Infiniband interconnects.


System Architecture

The Altix 350 implements the IP57 processor, but follows the IP41 architecture from the earlier Altix 3000 systems. A system can include from 1 to 32 processors in up to 16 modules or bricks, with one or two [NUMAlink] routers connecting them. Configurations up to 8 modules can be organized in a ring topology without a router.

Altix 350 systems can be composed of four different modules.

  • Base module
  • CPU Expansion module
  • CMPX Expansion module ("Cpu Memory Pci/pci-X")
  • NUMAlink 4 Router module (often labeled "NL4R")

However every system must have at least one base module, and any system with more than 8 modules must include at least one router.

Topologies

There are several different ways to organize the modules in an Altix 350 system.

  1. Simple ring
  2. Single-plane
  3. Dual-plane

CPU Options

The Altix 350 has been offered with a variety of Itanium2 processors. Examples have been seen with everything from 1.0GHz Deerfield units (1.5MB L3 cache, 62W, SL754) through 1.6 GHz Madison parts (reference needed), and several speeds in between.

Intel's later dual-core Montecito CPUs support the older bus signaling used on Madison CPUs, but at this time it is unconfirmed whether or not SGI ever released firmware support for these CPUs. In any case, such an application would be giving up many advantages of the Montecito design - DDR2 memory, PCIe support, and three times the FSB throughput compared to Madison.

Each Altix 350 base module will accommodate one or two Itanium 2 processors; other modules will accommodate 0, 1, or 2 processors.

Memory Subsystem

The Altix 350 supports registered DDR DIMMs with ECC, rated at 266MHz or faster, which are commonly referred to as PC2100, PC2700, etc. Three banks of four DIMMs each, or twelve slots total, provide up to 24GB of RAM per module. DIMMs of 512MB, 1GB, and 2GB are officially supported; it is unclear as of this writing if 4GB parts will work.

I/O Subsystem

Each module includes the following:

  • Front panel LCD display and buttons
  • Two NUMAlink ports
  • L1 USB Type B connector
  • L1 DB9 serial console port

Base module and CMPX modules provide two 64 bit PCI/PCI-X buses with two slots each, or a total of four PCI/PCI-X slots per module. Note that in base modules one slot is used for the required IO9/IO10 card. These buses can accommodate cards running at 33, 66, 100 or 133MHz.

Base modules and CMPX modules can optionally be fitted with an internal slim-line optical drive as well as two 3.5" Ultra160 SCSI hard drives, or two 3.5" SATA hard drives. The IO9 controller supports Ultra160 SCSI devices and includes an externally-accessible VHDCI port, while the IO10 controller supports SATA devices and includes a high density serial connector. Both controllers include a separate IDE interface to support optical drives.

Either IO controller provides one 1000baseT Ethernet interface.

The IO9 or IO10 must always be installed in Slot 1, the bottom slot in the base module.

Hard Drive Carriers

Intel-sr2300-drive-sled.jpg

The Origin/Onyx 350, Prism, Altix 350, and Altix 450/4x00 all use the same drive carriers. These are standard parts used by a number of manufacturers, notably Intel's SC5200, SRSH4, SR1300/2300, and SR1400/2400 and Sun's v60x and v65x servers.

The units carry "Assy A65278-00x" on a sticker, where the "-00x" may end in any digit though "-005" seems common. This assembly number can be useful when searching for parts in the Internet. According to SR1300/SR2300 support documents at intel.com, the Intel accessory part number for the drive carrier is FXX2DRVCARBLK, UPC code "7 35858 14621 0" and "MM #" 835853. However note that there is some indication that Intel may recycle these accessory part numbers in the FXX- form between different models.

As of early 2011 these drive sleds are commonly available on eBay for US$10.

Forum Threads About Origin 350 Drive Carriers

Add-On Options

Many PCI/PCI-X cards are available with support for Linux on Itanium.


Operating System Support

Linux

The Altix family was originally supported by SGI's Advanced Linux Environment, but eventually support was added to stock SuSE Linux Enterprise Server (SLES) or RedHat Enterprise Linux (RHEL) distributions for Itanium. As of mid-2013 SGI seems to only support SLES on any of their Itanium platforms, while both may be supported on their Xeon-based systems depending on model.

Several Altix 350 owners have reported success installing Debian Linux, and at least one has successfully built and installed Scientific Linux CERN 5.4 and CentOS 5.x through 5.9.

SGI offers the ProPack and SGI Foundation Software for Linux, which provides system management and performance monitoring tools beyond what ships with the base Linux distributions they support.

Troubleshooting and Repair

Engineers from SGI confirmed reports that a bug in older PROM versions would not work with the Linux 2.6 kernel. PROM version 4.43 is known to be compatible with SLES 11 SP1.

Replacement Parts

The Altix 350 uses a common Delta Electronics DPS-500EB modular power supply. While the DPS-500EB E variant is most commonly reported by 350 owners, others have successfully substituted DPS-500EB A units. However there is a known failure mode for earlier versions of these PSUs when run 24/7 for extended periods. See this thread for details and a fix.

Mixed PROM Versions

If you have modules with different versions of the PROM connected to each other it could cause problems. In some cases it could even prevent your system from completing the power-on self test successfully and reaching the EFI Shell. Not to worry though, forum member rosmaniac found a solution and shared it in this thread. This was the basis for a how-to posting in the Tips & Tricks forum.

Further Reading

Wiki articles:

Forum threads:

External References

Press Releases