jason andrade | 5 Jan 2004 21:55

Re: PERC3/DC, PERC4/DC Write Performance

On Mon, 5 Jan 2004, Richard, WhidbeyNet NOC wrote:

> Unfortunately, 8 MB/s does not even fully utilize a 100 Mb/s NFS
> network. When reading from and writing to the same RAID 5, writes are
> of course interrupted, and slowed to 4 MB/s. We had been planning to
> use the Gigabit NICs on the 2650, and the PV220S for backup, data, web
> and e-mail storage.

nod.  we're in a similar position with what we do in our architecture
in terms of using NFS (over gigE) and not wanting disk to be the
bottleneck.  we long ago decided to go with external raid controller
based storage to avoid the issues that are (generically) caused by
host based raid controllers (my not so great experiences with Digital's
KZPAC/PSC a long time ago :-).

> A Fiber/SAN configuration was the first thing our team examined. It
> would have cost $84,549 for two, fully redundant clusters connected to
> one CX200. Two redundant SCSI clusters, with two 220S enclosures and
> twice the storage, cost $35,270.
>
> We realized by going SCSI we would not reap the benefits of a SAN.
> However, even understanding that a SCSI cluster does not utilize cache,
> we could not have anticipated the throughput we achieved. Dell should
> inform customers of the extent their clustering solution impacts write
> speeds, if not before sale, then when a customer first calls for
> support on the issue.

you didn't have to go full SAN though it probably makes sense from
the BC POV.  you could have direct attached a lone CX200 to two
hosts in a cluster without needing switches.  and i agree with your
statement.  one of the other things to try is running Windows in
clustered mode.  i've seen some strange differences where Windows
'performance' was vastly superior to Linux on the same platform
which led me to believe some of the performance issues may not be
hardware implementation but driver/OS related.

> > did you consider asking dell for the PERC4/QC instead of the /DC ?
>
> The Quad-Channel cards are not compatible with clustering according to
> Dell:

bugger.

> There are many customers who have reported this issue before, on this
> list, others, and forums. We hope this helps those whose concerns met
> no conclusion, and are only asking that Dell set realistic expectations.

have you contacted your account rep who sold you the kit and asked
for your concerns to be escalated ?  dell have a pretty flat management
structure so your feedback shouldn't have to travel far.  though you
may have to be persistent to get past a lower layer.

regards,

-jason

_______________________________________________
Linux-PowerEdge mailing list
Linux-PowerEdge <at> dell.com
http://lists.us.dell.com/mailman/listinfo/linux-poweredge
Please read the FAQ at http://lists.us.dell.com/faq or search the list archives at http://lists.us.dell.com/htdig/


Gmane