Re: [Bacula-users] Performance with latent networks

On Thu, 2011-09-29 at 21:12 -0700, reaper wrote:

sounds like your sysctl.conf needs tweaking? have you tried something like this on both sides? 

 kernel.msgmnb = 65536 
 kernel.msgmax = 65536 
 kernel.shmmax = 68719476736 
 kernel.shmall = 4294967296 

 # long fat pipes 
 net.core.wmem_max = 8388608 
 net.core.rmem_max = 8388608

Yes, I tried to tweak kernel first with no effect. BTW wouldn't badly tuned network stack affect other programs such as iperf and ssh?

hi repear,

i'm going through a similar issue -- zurich, barca, copenhagen ...

iperf's results will differ dramatically depending on the tcp window size -- here's what i'm testing with:

server side (distant): /usr/bin/iperf -s -w 4096KB
client side (near): iperf -w4096KB -c distant.Server.IP.addr

(remember that iperf doubles the -w argument)

i have noticed that scp is not a good measure of throughput -- i do not know why. i use an openvpn tunnel between sites and loose about 20% of throughput due to the tunnel. check window size on distant machine (using wireshark) to verify that some upstream device is not changing it.

here's good site that discusses window size -- you can also google "long fat networks" for a detailed discussion; http://bradhedlund.com/2008/12/19/how-to-calculate-tcp-throughput-for-long-distance-links/

cheers

m

------------------------------------------------------------------------------
All of the data generated in your IT infrastructure is seriously valuable.
Why? It contains a definitive record of application performance, security
threats, fraudulent activity, and more. Splunk takes this data and makes
sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-d2dcopy2

_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users