LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: super slow speeds from director FIXED!

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: super slow speeds from director FIXED!
From: Roberto Nibali <ratz@xxxxxxxxxxxx>
Date: Fri, 23 Feb 2007 09:15:58 +0100
Matthew,

Well isn't this fan-damn-tastic. Turns out it wasn't our problem at all. We noticed that things started picking up speed around 3. We all assumed it was because I moved www to point to wwwdb1, thus bypassing the LVS.

 From our hosting provider, thePlanet.com:

After careful analysis of outbound throughput issues with various providers The Planet engineers determined that a gradual route table exhaustion was at the core of an increasing number of issues. It was determined that to repair this issue an emergency reload of the router dsr01.dllstx3 was required. At approximately 2:42pm CST we reloaded this router. There was a brief period of packet loss as the redundant assumed the load and at 2:51pm CST as the router re-converged. By 2:53pm CST all traffic was forwarding normally.

aehm, (cough) ... monitoring ... (cough), nice, they've solved the issue.

Amazing.

Yeah, what kind of SLA do you have with this provider?

For some reason I feel necessary to apologize to the list for breaking out the whip in my search for answers. Turns out it was outta my hand from the beginning.

No problem. But I'd like to pick up on your test case again. If you want to test bandwidth throughput and latency, your test (loading the picture and heuristically calculating the time spent downloading and rendering it via javascript) is not sufficient, because:

1. The size of the picture is too small regarding your network latency
2. The calculation is very sub-optimal
3. No statistically correct reproducible test possible

Thanks to all who offered help, most notably Joe, Graeme and Roberto. You guys rock!

I always tell people, sleeping over a problem helps. I'm glad I went to bed :).

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

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