[Spread-users] Tuning Windows XP for Spread
John Lane Schultz
jschultz at spreadconcepts.com
Thu Jul 12 12:39:18 EDT 2007
Doug Palmer wrote:
> On Wed, 2007-07-11 at 11:01 +1000, Doug Palmer wrote:
>
>> We've been running 4-5 spread daemons at High priority. However, we keep
>> on seeing the tok_hurry count go up and cases where one spread daemon
>> (apparently because it's the segment leader) suddenly develop a huge
>> number of s_retrans counts, which just keeps on going up. At the same
>> time, the spread deamons start growing in memory size and eventually
>> lock up.
>
> Further investigation reveals that the run-away s_retrans occurs when I
> send a message of size 1266 bytes. All other messages are <1k. So I
> suspect that something is going wrong in the message fragmentation.
>
> By switching around entries in spread.conf I can get the segment leader
> to shift. However, the run-away s_retrans occurs on the system that sent
> the message. So the segment leader idea is wrong.
>
> When should tok_hurry be sent and by what? At the moment, the count goes
> up on all daemons (from sptmonitor) every 2 seconds (well, 5 every 10
> seconds). I get the impression that this shouldn't be happening.
>
> What does s_ (or u_) retrans mean? This doesn't seem to be documented
> anywhere.
>
s_retrans stands for segment retransmission. This occurs when multiple daemons
within the same segment are missing the same packet.
u_retrans stands for unicast retranmission. This occurs when a single daemon is
missing a packet.
tok_hurry is when a daemon requests the token because it thinks it is lost.
If you are consistently getting tok_hurry's, which do occur by default every two
seconds, lots of s_retrans and u_retrans then you probably have faulty network
hardware somewhere in your setup and packets are getting dropped by the network.
I'm not sure if/how this relates to the runaway problem you've noticed with
pakcets bigger than 1K.
Cheers!
--
John Schultz
Spread Concepts LLC
Phn: 443 838 2200
Fax: 301 560 8875
More information about the Spread-users
mailing list