On 04/09/11 7:08 AM, Christopher Chan wrote:
>> In that case there's no real point upgrading. :-) The drives are U160's.
> The key is the backplane, not the drives. You should be able to get u320
> speeds if you have enough u160 drives
SCSI (parallel) before U320 had a weak spot. the command and status
phases of a SCSI command execution were stuck in asynch mode, only the
data transfer phase could use synchronous ('fast', 'ultra', etc).
Async SCSI is the old original 3-5Mhz mode, which on a wide bus means
6-10MB/sec. This wasn't a big problem with the original Ultra Fast
Wide 40 and Ultra 80 low voltage differential modes, but when SCSI hit
U160 it became a potential bottleneck, especially for high rates of
small block random IOPS on raids with more than a few drives.
IIRC (rusty neurons), a SCSI command takes 16 bytes or so, and a status
query takes 4-8. at the typical async 3Mxfer/sec, that means a command
takes like 5 microseconds. In 5 microseconds at U160, you could have
transferred 5*160 or 800 bytes, and a single status poll cycle takes
another 16-200 bytes, or pretty much 1000 bytes total, eg, for every IO
operation on the SCSI bus, you lose 1KB worth of potential data
bandwidth... So if you're executing 1000 IOPS spread across 4-6 drives,
you're losing at least 1MB/sec of your channel's bandwidth potential.
in reality its several times worse than this due to the additional
overhead of getting on the channel and the typical multiple polling
cycles required to complete a transfer, I've calculated its as bad as
10% under heavy IOPS random multiple drive raids
U320 supports synchronous command and status, so its way more than
double U160 speeds due to the elimination of this overhead.
Interesting side note, U320 controllers and U320 devices can support
synchronous command and status even if its running at U160 due to
backplane and cable loading issues.
thankfully, with FC, SAS, and SATA, these things are all history.