CUG.log Contents Back issues CUG SGI Contact us

T E C H   R E P O R T S

Operations SIG report

Service from SGI is presently evolving . . .

Dan Drobnis
Dan Drobnis,
former Operations
SIG chair


by Dan Drobnis

A big message for folks concerned with operating Cray products is that service from SGI is presently evolving.

Several new faces combined with many old friends combined to provide some sense of how the future is unfolding. Bob Brooks, new SGI vice president of Field Service, explained how SGI is moving to a single worldwide system of call centers ("I don't know how many call centers we currently have"), support centers, and defect tracking throughout SGI products. CRInform and SupportFolio will be merged in the second quarter of 1999.

Ken Coleman, new SGI vice president of Services, explained in a General Session how this worldwide system will be supported with a brand new information technology infrastructure, to be available in 12 to 18 months.

Some J90 sites with gigaring reported unusually low reliability. Buildable UNICOS source for J90 sites continues to be an issue. "Viewable" source does not address the issue of local mods and slow SPR response.


T90 reliability discussed

A birds of a feather (BOF) session on T90 reliability was well attended. Sites with large configurations are recording a considerably reduced mean time between failures (MTBF) from ones with small T90s:

 ObservedGoal
T94s1750 hrs2500
T916s370 hrs600
T932s 320 hrs 400

Dave Kieffer and several associates reported on what SGI has learned and is doing to improve T90 MTBF:

  • A process problem with all Emitter-Collector Dotted Logic chips that SGI/Cray assembles has been identified. Several parallel improvement efforts are underway, but sufficient time for accelerated life testing (about 1000 hours) will be needed to verify which are effective.

  • Some basic changes, such as improvements in cleanliness and screening tests, have already been made and chip reliability improvements noted, but not yet to a satisfactory level.

  • Improved chips will be introduced into manufactured boards and spares, but there are no plans for a wholesale swap of modules in the field.

  • Other changes, such as improvements in internal grounding, clock signal distribution, reduced operating temperature, and better connector cleaning and alignment procedures, are also being distributed to the field.

  • Spares inventory greatly increased, with more rigorous retesting of failed modules upon return.


Disk and tape technology examined

SGI's Mike Anderson provided his usual excellent survey of current and future disk and tape technology, particularly as it applies to Cray products. His summary:

The desktop drives the disk market because of sheer volume, but all drives are not equal! Quality and reliability still command a premium price. Where 18 GByte drives are now becomming common, 36 GByte drives will become so in '99. Product life is now only about 12 months before the next major technology advance.

SGI will utilize the 3.5" form factor through 2001. Increased data density on each drive makes transfer rate a potential bottleneck: A Model E I/O system with 79 GBytes used 41 DD-60 drives with a transfer rate of 840 MB/sec; a gigaring I/O with 80 GBytes uses 4 DD-309 drives with a transfer rate of 84 MB/sec. Fiber Channel Interface will replace SCSI as the high-performance choice.

Mike sees tape technology with another linear format war coming--STK Eagle vs. LTO (Linear Tape Open) from IBM/HPO/Seagate. Neither is backward compatible with 3480/3590 cartridges, even though the physical form factor is the same. Newer helical formats will allow capacities to 800 GBytes per cartridge and rates of up to 80-160 MB/sec.


See you in Minneapolis!

Due to the new SIG restructuring (see "CUG SIGs get a facelift" and "A guide to the new CUG SIGs"), the Operations SIG will now be called the Operations Focus Group. It will be chaired by Brian Kucic and reside under the new Computer Services "super" SIG.

Operations sessions at next year's CUG in Minneapolis are already shaping up: a report has been scheduled from the first SV1 site, and papers from several other sites with a variety of configurations are planned. Hope you can make it!

rule
Contents      Back issues      CUG      SGI      Contact us