> From: [email protected] [mailto:[email protected]]
> On Behalf Of Tom Limoncelli
> 
> Lastly, try to
> verify that the MTU on each router/device/whatever from the sender to
> the receiver is set to max (usually 1500).

You want your MTU set to 1500, ONLY if you're on the same network with each
other, or you're not doing any encryption, or all the routers in between
have an MTU higher than 1500.

The easiest way to see this problem is to establish a hardware-router-based
VPN between networks.  The endpoints don't know any encryption is happening,
so they're using 1500 MTU.  But in reality, the routers in between need to
encapsulate those packets to encrypt and send them, so every packet that's
already at the maximum size gets fragmented in half.  The easy solution to
this is to set the MTU of the endpoints a little smaller, say 1400.

But this MTU business won't create a discrepancy of 1.2gbps versus 6gbps.
The MTU business will only cause some extra latency, and a few percent
bandwidth overhead.  So I really don't think it's related to Andrew's
situation.

You can verify whether or not the router folks are right about the PSH flag,
by running a packet capture on your solaris & linux boxes.  If you're
getting a bunch of small packets, and you see the PSH flag is set...  Then
the PSH is probably causing the problem.  But be careful before trying to
disable it.  Depending on what protocol you're supporting, it *may* be
necessary or beneficial to keep the behavior as-is.

If you're doing something like a synchronous (notice, I did not say async.
I said a "sync") filesystem mounting, or something like that, then the
client might *need* to push the sync across the network to ensure it's
completed before starting the next packet.  If this is the case, you might
be able to improve performance at the expense of reliability...  Or you
might be able to improve performance without sacrificing reliability, by
using a different protocol.  Or, for example, if it's NFS with the sync flag
set, you might evaluate reliability implications and change to async mode.

etc.

_______________________________________________
Tech mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
 http://lopsa.org/

Reply via email to