SUN SPARCstorage Array Survey Summary

I received 23 responses to my SPARCstorage Array survey.

Note: This page is constantly under construction, and may not contain the latest information. Consult Sun technical support for definitive answers, and join the ssa-managers mailing list to participate in interesting discussions about the SSAs.

Disks

There is a firmware bug for the Seageate Barracuda 2.1 gig disks used in the SSA 102's. Make sure your firmware is 416 or higher. On May 3, 1996, Sun released patch 103451-01, with which you can upgrade to disk firmware revision 418.

Hot Spares

Most respondents had seen the problem where hot spares do not kick in. This is now a well-known bug. The hot-sparing algorithm is supposedly being totally revamped in the next version of Veritas Volume Manager, due in a couple months (Summer 1996).

There was some misinformation as to what exactly the hot sparing bug is.

• My description of how Veritas Hot Sparing works
• Rumors about a possible workaround

RAID5

A couple sites said that they saw data loss similar to ours using RAID5 under Veritas Volume Manager. This problem is still unresolved. We have chosen to move to a striped, mirrored configuration using Veritas (RAID 0+1), which sounds pretty stable from the responses I got (also, we had run it for a year before RAID5 with no problem). Now that our dissatisfaction with the SSA's has finally registered, Sun is "extremely interested in this issue" and wants to know more details to try to figure out where the bugs are.

According to my survey, hardly anyone is running RAID5 under Solstice DiskSuite. Those that are were running DiskSuite to begin with.

NVRAM

Many respondents said they were running with NVRAM turned on. Sun just found out about a serious bug in the NVRAM code which causes data corruption, especially with Solstice DiskSuite (SDS). Turn it off until you install patch 103290-02 or higher.

• Text of Sun's NVRAM Bug Announcement
• README.103290-02
• Postscript description of the NVRAM Bug from Sun

Prestoserve

One site had trouble until sun gave them a "supported configuration." They have been running fine with Prestoserve and SSA NVRAM turned on since March, 1996.

Performance

• 6 said performance was great!
• 11 said they performance was acceptable, or they got mixed results.
• 5 couldn't say, or forgot to answer the question.

We have found by learning the hard way that the particular layout of your volumes is very important to performance. RAID5 imposes several performance hits aside from having to write the parity information:

• It must compute parity information.
• Logging must happen on a separate spindle for acceptable performance. What a waste of a spindle, since the log only takes up a few megabytes.
• If you often update fewer than half of the stripe columns, RAID5 has to perform a Read/Modify/Write which can more than double the number of I/O's needed to change a couple blocks (read all disks, recompute parity information, write all disks).

Overall Satisfaction:

• 13 said they were satisfied with the SSA's and weren't really looking for another product.
• 9 said they were not satisfied and probably would not buy more SSA's.
• 1 did not have the SSA's long enough to have an opinion.

Last updated: Thu Jun 20 23:08:45 1996

Back to marg's SUN SPARCstorage Array Page