Jim,
That is exactly what we did when we wrote "data" to the drives. Let me add
that we measured the throughput at the host end of the SAN pipe. That is,
we used iostat under Solaris to watch kb/s numbers over the course of the
experiment. To recap, we saw:
All zeroes gave us a throughput to a single drive of almost 69MByte/sec.
Replace the zeroes with data and we only got 40MByte/sec. The same pattern
appears when writing to three drives simultaneously: zeroes add up to
~135MByte/sec; data to ~100MByte/sec.
There must just be issues with buffering; interrupt response times; and
other, higher level protocols that cause this behavior.
PaulW
www.pbnj-solutions.com IT Solutions That Stick
(216) 533-5708
> -----Original Message-----
> From: veritas-bu-admin AT mailman.eng.auburn DOT edu
> [mailto:veritas-bu-admin AT mailman.eng.auburn DOT edu] On Behalf Of
> jmcdon23 AT csc.com DOT au
> Sent: Saturday, April 26, 2003 5:44 AM
> To: Shafto, Eric
> Cc: veritas-bu AT mailman.eng.auburn DOT edu
> Subject: RE: [Veritas-bu] PCI bus and HBA throughput
>
>
> Hi
>
> Interesting but have you made allowance for hardware
> compression? Are you measuring the compressability rate of
> all those "zeroes"? Where is the bottleneck - at the
> tapehead or compression firmware? Try the same exercise with
> data that has been already compressed.
>
> Regards
> Jim McDonald
>
> --------------------------------------------------------------
> --------------------------
>
> This email, including any attachments, is intended only for use by the
> addressee(s) and may contain confidential and/or personal
> information and may also be the subject of legal privilege.
> Any personal information contained in this email is not to be
> used or disclosed for any purpose other than the purpose for
> which you have received it. If you are not the intended
> recipient, you must not disclose or use the information
> contained in it. In this case, please let me know by return
> email, delete the message permanently from your system and
> destroy any copies.
> --------------------------------------------------------------
> --------------------------
>
>
>
>
>
> "Shafto, Eric" <Eric.Shafto AT drkw DOT com>@mailman.eng.auburn.edu
> on 04/26/2003 12:42:50 AM
>
> Sent by: veritas-bu-admin AT mailman.eng.auburn DOT edu
>
>
> To: "'Paul Winkeler'" <pwinkeler AT pbnj-solutions DOT com>,
> "'Vijay Korde'"
> <vijay_korde AT hotmail DOT com>
> cc: veritas-bu AT mailman.eng.auburn DOT edu
> Subject: RE: [Veritas-bu] PCI bus and HBA throughput
>
>
>
> It would be interesting to see whether the situation
> improves noticeably with multiple HBAs. Does anyone have
> some data to share? -----Original Message-----
> From: Paul Winkeler [mailto:pwinkeler AT pbnj-solutions DOT com]
> Sent: Friday, April 25, 2003 8:57 AM
> To: 'Vijay Korde'
> Cc: veritas-bu AT mailman.eng.auburn DOT edu
> Subject: RE: [Veritas-bu] PCI bus and HBA throughput
>
>
> Hi Vijay
>
> Here is what we found using two other brands of PCI Bus
> 2Gbit HBA's in SunFire 880's going to T9904B's:
> 1) To a single drive using 2Gbit HBA's, observed peak
> ~40MByte/sec, regardless of 33 or 66Mhz bus
> 2) Writing 3 streams to 3 drives across the same 2Gbit HBA,
> observed peak ~100MByte/sec regardless of 33 or 66Mhz bus
> During all tests source data was fully cached in RAM
> (machine had 8GByte) and no other processes running (machine
> had 8 CPU's)
>
> Wconcluded that the bottleneck was likely in the handling of
> the SAN protocols across the HBA. Our basis for this
> conclusion was this:
> - Writing zeroes (/dev/zero) to a single T9940B drive
> yielded a rate of 69MByte/sec which is basically the upper
> limit at which StorageTek claims the drive can take data in;
> so we know it can go that fast.
> - Pushing zeroes to 2 drives simultaneously can be done at
> an aggregate rate of ~130MByte/sec on the 66MHz bus, closer
> to ~120MByte/sec on the 33Mhz bus.
> - Going to 3 simultaneous drives with zeroes yields rates of
> ~138MByte/sec and 129MByte/sec respectivel for the 66MHz and
> 33Mhz buses. In other words, the bus speed gave the 66Mhz
> bus a slight edge but hardly worth bothering. It would be
> very interesting to see what the results look like when you
> use multiple HBA's simultaneously...
>
> My advice: don't hang more than 2 drives of a single HBA.
>
> PaulW
> www.pbnj-solutions.com IT Solutions That Stick
>
>
> If you have received this e-mail in error or wish to read our
> e-mail disclaimer statement and monitoring policy, please
> refer to http://www.drkw.com/disc/email/ or contact the sender.
>
>
>
>
> _______________________________________________
> Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
> http://mailman.eng.auburn.edu/mailman/listinfo> /veritas-bu
>
|
