Feed problem

Brennon Church ctsbac at cts.com
Thu Oct 19 19:33:53 UTC 2000


Okay, here is yesterday's INND timer report:

article cancel 		00:41:52.051  2.9%   133677  5.677  18.792 257.067 
article control 	00:41:35.093  2.9%   132858  5.723  18.780 257.133 
article link 		00:00:00.000  0.0%        0  0.000   0.000   0.000 
article write 		06:13:25.766 25.9%   941894  9.085  23.788  39.102 
history grep 		00:09:15.291  0.6%    54927  0.345  10.110 187.208 
history lookup 		01:59:58.795  8.3%  6022207  0.274   1.195   6.768 
history sync 		00:01:34.792  0.1%   108916  0.000   0.870  41.283 
history write 		05:27:47.534 22.8%  1087561 10.378  18.084  55.422 
idle 			01:08:03.738  4.7% 24779260  0.081   0.165   1.097 
overview write 		00:00:00.000  0.0%        0  0.000   0.000   0.000 
perl filter 		05:20:25.811 22.3%  1006476  6.623  19.102  31.227 
site send 		00:01:14.328  0.1%   941894  0.021   0.079   0.606 

TOTAL: 23:59:29.006 	21:45:13.199 90.7%        -      -       -       - 

So it would appear that the three points of concern are article write,
history write, and perl filter.  I've disabled the perl filter for the
moment.  I'm guessing that the article write is also anomalously high.

Does anyone see something I'm missing perhaps?  I'm going to let the system
go for a day as-is without the perl filter.  After that, I'll probably
start playing around with the drives and such to see if I can improve the
write time.

-- Brennon Church

On Mon, Oct 16, 2000 at 11:15:15PM +0200, Fabien Tassin wrote:
> hi,
> 
> can you post a typical "INND timer" report (from an innreport daily report)
> for your server ? this often helps to locate problems..
> 
> According to Brennon Church:
> > 
> > Hello,
> > 
> > I'm trying to trace down a problem where one of my news system is unable to
> > accept a full news feed (150G/day +).  I'm running the 20000715 snapshot of
> > inn-2.4.0 on a FreeBSD 4.0 stable box.  I'm using cycbuffs with no
> > overview.
> > 
> > In watching the output from netstat, I can see that the incoming sockets
> > are spending most of their time full or near full.  That would seem to
> > indicate to me that the local system (my machine) isn't able to pick up the
> > data fast enough.  I'm thinking this would have to do with one of three
> > things, processor, ram, or hard disk.
> > 
> > The processor is a PIII 800 at 133Mhz, so it should be plenty fast enough.
> > Additionally, top shows that it usually hovers around 40% to 50%  idle, with 
> > innd taking up an average 25% to 30%, and innfeed taking up between 2% and 
> > 5%.
> > 
> > The ram is all running at 133mhz, and I have a total of 768M of it.  Top
> > tells me that innd is running at ~140M total, and ~60M resident.
> > 
> > There are two drives in the system, one 4gig boot drive, and a 50gig spool
> > drive.  The system is transit only, so retention isn't an issue.  Watching
> > a systat iostat, I see that the transfers per second go between 50 and 100,
> > and the MB/s is about 1.5 to 1.7.
> > 
> > Does anyone have any experience with a similar problem?  Or are there any
> > ideas on what I might be missing in this whole thing.  I've done a
> > significant amount of testing, so let me know if I need to expand upon
> > anything I've mentioned here.
> > 
> > On a side note... What is about the average size for a full news feed
> > nowadays.  I've managed to get up to 152Gigs in a day, but I'm usually
> > hovering around 140 or so.
> > 
> > --Brennon Church
> > System Administrator
> > CTSnet Internet Services
> > 
> 
> -- 
> Fabien Tassin -+- fta at sofaraway.org



More information about the inn-workers mailing list