LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: super slow speeds from director

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: super slow speeds from director
From: Roberto Nibali <ratz@xxxxxxxxxxxx>
Date: Thu, 22 Feb 2007 21:08:00 +0100
Morning All,

Evening sir,

 >> Since our local office is
having problems, I'll tcpdump an office d/l and then one from home. Will post when I get them.

reasonable.

I have posted 4 tcpdumps, all grabbing the speedtest.html page:

What OS are the clients running?

from office, www: (super slow)
    http://www.omnovia.com/test/cluster_dump_work.txt

from office, wwwdb1: (normal speed)
    http://www.omnovia.com/test/direct_dump_work.txt

No handshake visible, please capture the whole TCP flow.

from home, www: (normal speed)
    http://www.omnovia.com/test/cluster_dump_home.txt

from home, wwwdb1: (normal speed)
    http://www.omnovia.com/test/direct_dump_home.txt

All dumps were done on wwwdb1 when accessing. Hopefully you guys can make heads/tails out of those dumps cause I don't know what I'm reading.

Hmm, could you add at least one '-v'? To me it's not enough information to figure out what's going.

I did notice that in both cases, going directly to wwwdb1 used about 200 packets and both cases of using the cluster, around 100.

Now this is interesting! What strikes me as odd in your dumps is the client address with a trailing '0', aka 70.241.140.0. I'm not sure how TCP stacks react to this, but could you for the sake of a clean test conduct change your IP address of the test client to something between (including) 70.241.140.1 and 70.241.140.254? Also note, that testing from within the cluster does not give you proper results due to the asymmetric routing.

Regards,
Roberto Nibali, ratz
--
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq' | dc

<Prev in Thread] Current Thread [Next in Thread>