On 3-3-2010 10:10, simon wrote:
Hi there
I am experimenting with using memcache for storing PHP sessions in a
www cluster environment. But due to latency in network, I have noticed
slight performance drop in application as opposed to just direct
access.
Has anyone played with changing MTU on their network card to minimize
latency? Hence get faster response?
http://www.speedguide.net/read_articles.php?id=111
MTU dictates the *maximum* packet size, not the default one. I.e. if you
have to send just 100 bytes, a packet would be 140 whether the MTU is
500 or 1500. If you have to send 6000 bytes, however, you'd end up with
12 vs 4 packets. And thus the total latency for all packets combined
would be larger for the small-mtu-case.
There may also be additional latencies apart from the transmission-time
for the bytes. I.e. switches, nics and network stacks in your OS may
(will) add latency. And apart from that, there may be some ACK's in
between for some of those packets and there is a probably a larger
change of retransmissions due to the increased amount of packets. So the
above difference with data-transfers larger than your MTU is amplified
by all those things.
If your network has - at some point - a situation where it can only
transfer a MTU of 500 bytes, it may be a good idea to have all MTU's at
that path have only 500 bytes, to save on fragmenting-overhead. But in
most datacenter-LAN's thats not a very common scenario afaik.
You may actually be better off increasing the MTU (enabling jumbo
frames) rather than decreasing it, if you want to decrease latency.
All depends on your set-up of course, but I can't imagine decreasing the
MTU will help you to improve the latency.
Best regards,
Arjen
