RE: 10Gbps latency

Keep us in the loop.

Most likely the servers are setup poorly.... however can we also confirm -

The blade servers themselves "home run" back to a big Cisco switch (and do not use a little in-blade forward deployed small switch?)

-----Original Message-----
From: marc abel [mailto:marcabel_at_gmail.com]
Sent: Monday, July 11, 2011 11:11 AM
To: Joseph L. Brunner
Cc: ccielab_at_groupstudy.com
Subject: Re: 10Gbps latency

Thank you Joseph,

1. Not trying to access storage. this is actually just interconnect
traffic. The replication traffic is fine. The way it was explained to
me is that each node in the RAC queries the other nodes to see if they
have data cached before retrieving it themselves. This is where we are
seeing the compounded wait times.

2. Not positive, I will check.

3. Of course not, we have a "prove it's not the network first" policy
in effect here. All kidding aside, The servers seem to be performing
fine and within bounds but I have no way to know/validate what has
been done on the database side.

On Mon, Jul 11, 2011 at 9:59 AM, Joseph L. Brunner
<joe_at_affirmedsystems.com> wrote:
> Hi Marc,
>
> I face these debates/questions on almost a daily basis so I feel for you man...
>
> Can you confirm the following:
>
> 1. Are you trying to access iscsi storage or is this simply oracle replication traffic?
>
> 2. Are you using the very latest nic drivers on each server?
>
> 3. Can you confirm disk queue, raid level, raid performance and other NON-NETWORK related things have been checked and blessed by the server priest ordained at your location?
>
>
> Thanks,
>
> Joe
>
> -----Original Message-----
> From: nobody_at_groupstudy.com [mailto:nobody_at_groupstudy.com] On Behalf Of marc abel
> Sent: Monday, July 11, 2011 9:48 AM
> To: ccielab_at_groupstudy.com
> Subject: OT: 10Gbps latency
>
> Hello all,
>
> I have an issue where we have a 10gbps backbone, and our blade chassis
> are also 10gbps however the actual blades themselves are only 1gbps.
> This links are no where near saturated. Even the 10Gbps uplinks rarely
> carry over 50Mbps of traffic.
> We are having some performance issues with Oracle RAC, and the DB guys
> are of course trying to blame the network. They are claiming that
> upgrading the blades to 10Gbps will solve all our issues. I have
> explained that we are not saturating the links, we are seeing no
> discarding/queuing/etc, but they insist that we will see improvements
> in latency by upgrading the blades. They are basing all of this
> anecdotally on the fact that one of the DB developers ran into this
> at his previous employer and that is how they solved it.
>
> Most of the Cisco spec sheets I have found do show an improvement in
> latency when comparing Multiple 1Gbps links to 10Gbps, but seem to
> make the assumption that some of these links are saturated. This makes
> sense, but what about in the absence of much traffic, is a 10Gbps link
> "faster" than a 1Gbps link?
>
> I remain unconvinced that it will make a bit of difference, but would
> like to hear other opinions.
>
> Thank you,
>
> Marc
>
>
> Blogs and organic groups at http://www.ccie.net
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html

Blogs and organic groups at http://www.ccie.net
Received on Mon Jul 11 2011 - 15:14:38 ART