Skip to contentSkip to conversation

Adaptec AIC-79xx HostRaid

DISCUSSION Eric Jones 2 participants 2 messages May 30, 2005

lore
Share & tools
lore

Eric Jones

author
#0 lore raw
Quoted 5 lines
Quoted 3 lines
Quoted 3 lines

We have acquired some IBM xServers which have an integrated raid controller based on the Adaptec AIC-79xx U320 SCSI controller (called HostRaid).

As far as I know, it is software raid done much closer to the hw than the linux sw raid (md).

Actually, there are two possible configurations, depending on HW configuration...

For example the IBM x346 server has an onboard Adaptec 7902 SCSI controller chip for the U320 backplane devices (max-6 devices). This chip uses the standard aix79xx driver for normal SCSI functions. But, for extra $$$, you can also buy the IBM feature "ServeRAID-7k" which is a small DIMM-sized plugin module that adds XOR parity generation and a small cache.

http://www-132.ibm.com/webapp/wcs/stores/servlet/ProductDisplay?productId=8735579&storeId=1&langId=-1&catalogId=-840

Adaptec calls this a "zero-slot" RAID solution, and it requires a special binary driver to recognize and handle the array functions.

Most people seem to order this bit with the initial kit, but then rip it out when they realize it needs a binary driver to support the RAID functions. The Linux SW-RAID seems to provide better IO performance, and you can still see the individual sdx devices on the SCSI bus. This makes it easier to monitor device performance and health (ie. smartd, etc), vs the Adaptec "Hide everything behind the controller" approach with the ServeRAID-7k configuration.

Eric

#1 lore raw
Quoted 8 lines

But, for extra $$$, you can also buy the IBM feature "ServeRAID-7k"
which is a small DIMM-sized plugin module that adds XOR parity generation and a small cache.

http://www-132.ibm.com/webapp/wcs/stores/servlet/ProductDisplay?productId=8735579&storeId=1&langId=-1&catalogId=-840

Adaptec calls this a "zero-slot" RAID solution, and it requires a special binary driver to recognize and handle the array functions.

I would be very suspicious about the claim that the xor stuff is done in hardware and not in the binary part. I'm actually surprised that a linux friendly company like IBM actually gets involved with binary drivers like this though.... a bit disappointing.

but yeah this sounds like a really bad purchase.