adaptec 2120S report

[Scott Long]

Andrew Snow wrote:
> 
> After my failure to get the Serveraid-5i working, I have put in an 
> Adaptec 2120S. This card has very similar features to the Serveraid 
> card, with the addition that it actually works in FreeBSD.
> 
> After a couple of days of testing I can recommend this card.  I have 
> 6x36GB IBM 10krpm drives running in RAID 50 configuration with FreeBSD 
> 5.1-REL.  The only downside was I cannot get it working at Ultra-320 
> with this IBM x345 hotswap backplane, but it is working great at Ultra-160.

What tool are you using to report the speed?  I haven't tried the IBM
backplanes much, but it wouldn't surprise me if it didn't support
packetized mode and therefore not support Ultra320.

> 
> With my configuration, a quick test of file system sequential access 
> speed shows reads at 47mb/s and writes at 27mb/s.  The good news is, 
> while the array is degraded and rebuilding, read is 10mb/s and write is 
> 6mb/s which is still quite respectable. The box still feels fast enough 
> while the array is rebuilding. The downside is rebuilding the RAID5 
> containers took 5 hours - each.
> 
> Two problems with this controller in FreeBSD 5.1:
> 
> 1. FreeBSD won't boot while the array is rebuilding. It hangs after 
> detecting aacd0.  I did not try waiting 5 hours to see if it continued 
> after the array rebuild finished.
> Workaround is to only hot swap drives while FreeBSD is running, and do 
> not reboot until rebuild is finished.  Alternatively, boot into the 
> controller BIOS and wait for rebuild to finish there.
> 

This should not happen, and I cannot reproduce this locally.  What
circumstances surrounded the build?  Was it initiated by the BIOS?
Would it be possible to enable 'bootverbose' and send me a transcript
of the boot log?

> 2. I could not get SMP kernel to work - its a Xeon 2.6ghz supporting 
> hyperthreading. SMP worked without the controller, but hangs during boot 
> with the controller.  Luckily this is not a big problem for us as we 
> only need a single CPU and hyperthreading is of minor/dubious advantage.
> Workaround is to simply not use an SMP kernel.

Again, this is quite serious.  I've fixed several bugs since 5.1 was
released that could contribute to this.  However, it's another case of
not being able to reliably reproduce the problem locally.  Would you be
able to test a very recent 5-current snapshot?

> 
> Other than that, it seems to be working very nicely.  I look forward to 
> seeing the Serveraid / ips driver issues resolved in 5.2...

It's on my TODO list.  I apologize for not being responsive, I've been
quite buried under other projects.

Scott