Memcache connection errors

Jacques Caron jc at
Wed May 24 14:38:53 UTC 2006


At 15:29 24/05/2006, Kamran Nisar wrote:

>got zero tx/rx errors...

Did you check the switches as well? I guess you have everything 
correctly set to full duplex? What kind of switches are you using?

>even the ping returned 0% packet loss, though there was sum random 
>sharp jumps in repsonse time

0% or packet loss or 0 packets lost? In some situations you may have 
a very low packet loss (under 1%) that is still enough to generate 
some random timeouts, so make sure you check that you received 
exactly as many packets as you sent. Also the increases in response 
times mean you have some kind of network issue, probably a link or 
switching matrix maxing out somewhere.

You may also want to do a local test (httpd and memcache on the same 
box) if at all possible, to completely isolate network issues vs 
other types of issues. I suppose you don't have any rate-limiting 
filtering rules anywhere? If you can't run the httpd on the same box 
for whatever reason, I guess a short perl (or whatever other 
language) script running GETs on the memcached server (running on the 
same box) in a tight loop while you're running your benchmark will 
let you know whether the issue is local or not.

What's the load (CPU user/system), and memory usage (RAM, swap, 
pageins/pageouts) during your benchmarks? Is there anything else 
running on the memcached server?


More information about the memcached mailing list