Henrik Nordstrom wrote:
>
> Ok. Try this updated patch which tries to take ICP status into account
> when selecting parents based on statistical RTT. This is done by
> ignoring peers wich replied with a "odd" ICP status code on the last ICP
> query.
Thank you for the new patch - looks good.
The JANET national cache (or at least, two of the three systems there that
we use as parents) seem to be having even worse problems today - both the
systems at Manchester are now failing to accept TCP connections most of the
time and are responding with ICP_MISS_NOFETCH for misses. Useful for
checking the latest patch, though!
Except briefly during startup and after a reconfigure, the number of
TIMEOUT_FIRST_UP_PARENTs is very small and so is the number of
TIMEOUT_FIRST_PARENT_MISS fetches routed to parents that are behaving oddly.
For example, counts for the last 2000 access.log entries with PARENT or
SIBLING in the status, I now see
284 FIRST_PARENT_MISS/cam2.sites.wwwcache.ja.net
58 FIRST_UP_PARENT/cam0.sites.wwwcache.ja.net
1 PARENT_HIT/cam0.sites.wwwcache.ja.net
60 PARENT_HIT/cam2.sites.wwwcache.ja.net
2 SIBLING_HIT/wwwcache.damtp.cam.ac.uk
2 TIMEOUT_FIRST_PARENT_MISS/cam0.sites.wwwcache.ja.net
1586 TIMEOUT_FIRST_PARENT_MISS/cam2.sites.wwwcache.ja.net
7 TIMEOUT_FIRST_UP_PARENT/cam0.sites.wwwcache.ja.net
cam0 is the first parent in the list; cam0 and cam1 (the latter normally our
fastest parent) are both rejecting most TCP connections and returning
ICP_MISS_NOFETCH (with a few ICP_HITs as well from cam0). cam2 seems to be
behaving normally *but* is the slowest parent, normally the worst choice -
so it's good to see that the patch is recognising that in spite of slower
RTT, it should be used preferentially.
The high number of TIMEOUT_ entries is presumably because cam0/cam1 are
often responding late, if at all.
> (this time structs.h is changed, so don't forget to make clean)
Thanks for the warning!
John
-- University of Cambridge WWW manager account (usually John Line) Send general WWW-related enquiries to webmaster@ucs.cam.ac.ukReceived on Tue Jul 29 2003 - 13:15:58 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:07 MST