In [UDP packet loss in the stack] mail:
http://marc.theaimsgroup.com/?l=linux-net&m=103846662602797&w=2
Dheeraj told me to increase the memory
sysctl -w net.core.rmem_max=524288
sysctl -w net.core.wmem_max=524288
sysctl -w net.core.rmem_default=524288
sysctl -w net.core.wmem_default=524288
I did this at the UDP client side and that works. Altough i still have
packet loss (by streaming up to the max). Looking in the network dump i see
now and then packet loss (approx. 40 packets of 1500 bytes is ~~ 64K hmmm)
And this always happen when the server do an arp request (for the UDP
client). Can this be explained? Is the UDP server side of out buffers and
frees it some UDP buffer to get memory for other traffic like arp?
Oke, i tried to increase the buffers at the UDP server side.
sysctl -w net.core.wmem_max=524288
sysctl -w net.core.wmem_default=524288
This become a drama. Now i loss up to 75% of my packets!!!!
Looking at the logging at the client site it looks like it losses also
chunks of 40 packets.. or more
Q] Does anybody know why we get severe packet loss when we INCREASE buffer
space at the UDP server side?
Looking at ifconfig you can tweak the txqueuelenght (currently 100) but it
doesn't seem to influence the behaviour when increasing this to 512 or what
so ever...
Anybody know why we already lose packets in the stack? What is the policy of
the linux kernel with UDP packets when resources become scarce?
Hoping somebody can enlight the dackside of the networkingstack...
Thnax in advance,
Andre
_________________________________________________________________
MSN 8 with e-mail virus protection service: 2 months FREE*
http://join.msn.com/?page=features/virus
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
