More on -- Severely delayed (15s) ACK on inbound NNTP packets cau sing hung channels
brett.wyer at telecheck.com
Wed Oct 6 13:14:22 UTC 1999
My sincere apologies. You'd think I didn't do tech support...
I'm running Digital UNIX v4.0E and using the storage API -- cnfs method.
I've got six channels going against Sprintlink and being fed by a T1. I've
checked the usage and it isn't peaked. We're sitting at an average of about
800Kbps. The really strange thing is that when I pull up history from
shortly after I converted to cnfs, we were taking 95% of the news fed to us
without a problem. Now we're dropping over half and can't even keep a T1
Additionally, I've checked disk queues and CPU utilization. INN never
consumes more than about 30% of a CPU and disk queues never exceed 1-2 I/O's
outstanding in an 8-member stripeset.
A little more information:
Oct 6 08:10:42
Percent busy: ( 300003 - 212564) / 300003 = 29.1%
Time per article write: 2833 / 258 = 11.0 ms
Percent time spent writing art: 2833 / 300003 = 0.9%
Time per history write: 4051 / 350 = 11.6 ms
Percent time spent writing his: 4051 / 300003 = 1.4%
Time per history lookup: 451 / 704 = 0.6 ms
Percent time spent hist lookup: 451 / 300003 = 0.2%
More information about the inn-workers