[CentOS] TCP flow latency graphs

JohnS jses27 at gmail.com
Tue Sep 14 14:47:01 UTC 2010


On Mon, 2010-09-13 at 15:28 +0200, przemolicc at poczta.fm wrote:
> Hello,
> 
> we have one application which gets some data from our database but
> to print just one result it connects many times to the database over WAN.
> Of course there are some performance related problems with this type of work.
> I'd like to analyze (not necessary online) its tcp flow especially its
> latency (using port redirect on switch).
> Do you know any software (similar to tcpdump or so) which will let me 
> graph/paint latency between each connection and each tcp flow ?
> 
> Regards
> Przemek
---
I say talk to your DB Admin | Hit the Db once and not many times
(connection), there's your problem at least one.

Latency starts at the Application Layer then on down the Hardware then
the Network.  You can throw a lot of money away also on it.  There are
kernel options for networking latency (see the kernel pages docs).
Although you want improve it hardly any with the mainline kernel.  The
WIRE is the Killer and App priority.

To start with you are going over a link you have NO control of so your
really stuck unless you have DEEP Pockets.

Any way one of the best tools to use is Netperf [1].  You can build your
own rpm from the gz package.

What's the latency now???


John


[1]. http://www.netperf.org/netperf/
a loopback
netperf -c
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to foomahine
(127.0.0.1) port 0 AF_INET
Recv   Send    Send                          Utilization       Service
Demand
Socket Socket  Message  Elapsed              Send     Recv     Send
Recv
Size   Size    Size     Time     Throughput  local    remote   local
remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % U      us/KB
us/KB

 87380  16384  16384    10.00      3002.26   100.00   -1.00    2.729
-1.000 




More information about the CentOS mailing list