High TCP segment retransmit rate

Linux and Unix-Like Linux
1

Hi All,

we have some 20 CentOS hosts which are experiencing high TCP segment retrasmit rate. I checked with network team if it is something to do with switch, they say there is no latency or errors.
This is a issue now because the sybase quiries from the DB instances on this host are getting slower.
Can any one let me know how to reduce this TCP segment retransmit rate?

Thanks,
RK

2

there are some kernel options for ipv4 & ipv6 to set 'retrans_time'.
don't know if it works setting it to a higher value...

3

Hi daWonderer,

Changed this to 200 from 99 and restarted eth, but still didnt work.
Any other pointers please?

Thanks,
RK

4

Check the value to see if you get back out what you put into it.

I don't think you'd need to restart eth for this. That might even undo your changes.

5

Hi Corona,

Still no go...... :frowning:
Interestingly I found that this increases over a period of 4 hours 12:00 to 16:00 EST. I happen to see if any switch issues - no. any db/application traffic increase-no

So any one needs more info to dig through?

-RK

6

Did you get the same value out of it that you put into it?

7

Yes I reverted the value back to 99. and still not working.