more RAM + faster disk -> slower box?!

Mark Blackman mark at blackmans.org
Thu Nov 29 13:30:47 GMT 2001


Ok. fair enough. The excess CPU load (in cases of IO problems) can come 
from badly written code which is usually short-lived.

Say it spins (polls) waiting for a lock to come free, which normally comes
free in .001 seconds, but because of IO problems hangs around spinning
on a lock.

Obviously, this is completely avoided with a blocking flock.

The only other thing I can think of is to check your filesystem
settings, although I suppose you would see busy disks if that
were the problem. 

In any case, you're right and your duplex settings are fine, so thats
nothing to do with it.

Our news peers tend to run at about 2-6Mbytes/sec (network and disk)
continously, but I don't know what the server tends to run at. Although,
I believe its less than that.

Anything suspicious in /var/log/messages?

- Mark

> Hi Mark,
> 
> > Checked the duplex settings on both the switch and the NICs?
> 
> Not since swapping the disk and adding extra memory, but I'm not seeing any 
> errors, the switch is locked to 100/full for that port, and the NIC's happy.
> 
> bash# uptime; netstat -I fxp0; ifconfig fxp0
> 12:54PM  up 17:33, 1 user, load averages: 5.36, 4.81, 4.48
> Name  Mtu   Network       Address            Ipkts Ierrs    Opkts Oerrs  Coll
> fxp0  1500  <Link#1>    00:d0:b7:09:0e:f9  6078367     0  3865296     0     0
> fxp0  1500  manlan.mcc.ac godfrey.mcc.ac.  5459169     -  3864943     -     -
> fxp0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
>         inet 130.88.203.19 netmask 0xffff0000 broadcast 130.88.255.255
>         ether 00:d0:b7:09:0e:f9 
>         media: Ethernet 100baseTX <full-duplex>
>         status: active
> 
> > What kind of network throughput are you seeing (netstat -w 1)?
> 
> bash# netstat -w 1
>             input        (Total)           output
>    packets  errs      bytes    packets  errs      bytes colls
>        146     0     104412        103     0       5132     0
>         93     0     140850         54     0      14034     0
>         74     0     113765         54     0       7579     0
>         90     0     239803         60     0      10825     0
>        321     0      31797        212     0       3385     0
>         16     0       2139         13     0       2999     0
>         20     0      48359          9     0       3284     0
>         16     0     233811          7     0      70719     0
>         60     0      30900         41     0       2303     0
>        211     0      53243        139     0       3210     0
>         41     0      27158         32     0       1622     0
>         63     0     175736         37     0       6522     0
>        183     0       2587        110     0        804     0
>         19     0       2442          3     0       9735     0
>         23     0      16393         19     0       2605     0
>         30     0     244078         20     0       9886     0
>        144     0     307949         89     0      11302     0
>        236     0     160180        151     0       8498     0
>        157     0     117371         95     0       5210     0
> 
> ... but I don't know enough about the _expected_ traffic to interpret that.
> 
> > I've found duplex mismatch to cause me more problems than pretty
> > much anything else.
> 
> Sure, BTDT, hence the "config locked" ports in the switch. 
> 
> But I still can't explain why my supposedly IO-bound processes are spending a
> ll
> their time burning CPU, when all I've done is add memory...
> 
> Jesus, sorry, I sound like Joe Luser here, don't I -- "but all I did was to
> delete that .DLL thing". :-)
> 
> Thanks for the suggestion, but I don't think it's that. :-(
> 
> Ian.




More information about the Ukfreebsd mailing list