OpenVZ Forum: Support » Understanding user

Home » General » Support » Understanding user_beancounters: tcpsndbuf

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Understanding user_beancounters: tcpsndbuf [message #27849]

Fri, 29 February 2008 09:02

ittec
Messages: 41
Registered: February 2008

Member

Hi again

Since I modified some values of privvmpages on my Linux OpenVz server with your help, no more problems happens with this parameter.

Today I need to explain to you similar trouble, this time with tcpsndbuf. I read the http://wiki.openvz.org/Tcpsndbuf#tcpsndbufproperly thread in OpenVz wiki and I did the calculation about tcpsndfbuf formula:

$http://wiki.openvz.org/images/math/2/e/3/2e3d5c634d4612a3dda5744d4b8671ba.png$

Current user_beancounters values:

tcpsndbuf:
barrier -> 5.777.926
limit -> 12.331.526
fail -> 455.979.976

numtcpsock:
barrier -> 1.600
limit -> 1.600

I think the number of fail is very high. Now with this values Im going to follow the formula that appear in wiki web:

limit - barrier >= 2.5 KB * 1600

12.331.526 - 5.777.926 >= 4.000
6.553.600 >= 4.000

Is right, isn't it? So my first interest is about understand why I have a high number of tcpsndbuf. If I think in formula and I did properly, couldn't happen bad performance on this VE. But the true is the traffic graphics show me bad operation.

Can someone bring me some light here ? Confused

Thanks a lot

Report message to a moderator

Re: Understanding user_beancounters: tcpsndbuf [message #27926 is a reply to message #27849]

Mon, 03 March 2008 11:52

xemul
Messages: 248
Registered: November 2005

Senior Member

Quote:

So my first interest is about understand why I have a high number of tcpsndbuf.

High tcpsndbuf means that there are many active (i.e. passing data) TCP connections on the node.

Quote:

couldn't happen bad performance on this VE.

Small tcpsndbuf may cause a small performance degradation even when there's one TCP connection on the whole node. You may increase the one and check.

This limit was designed not to allow a VE users consume too many kernel memory by sending large amount of TCP packets. The difference between limit and barrier is used to provide a minimal speed for TCP connection when all the sockets are active.