Re: Linux Tuning

Stuart Lynne (sl@gateway.fireplug.net)
20 Apr 1999 18:07:00 GMT


In article <199904200408.AAA04596@adder.cs.Virginia.EDU>,
Greg Lindahl <lindahl@cs.virginia.edu> wrote:
>> There was some talk about system call accounting last week. The cleanest
>> place to add this is in assembly code which I'm not qualified to modify.
>> (I don't find the number of system calls a particularly interesting number,
>> but it should be there for correlation purposes.)
>
>You can get this for a given process by running strace -c. I'm not
>sure if there is much to be gained from putting it in the kernel.
>
>There *is* some interesting stuff to be added to the kernel for
>performance monitoring, but I would parallel the tuning
>discussion. Two examples:
>
>1) Resource exhaustion. If you get more than 300 packets in 1/100 of a
>second, you overflow the backlog queue. You can raise this by messing
>with /proc/sys/net/core/netdev_max_backlog, but how would you know to
>do this? One answer is to have the kernel print warnings when the
>queue fills and packets are dropped. These warnings should not appear
>too often, i.e. once every 20 minutes along with the # of packets
>dropped. The *bsd's are pretty good with this kind of thing.

Just add two counters, one to count the number of times the queue was filled
to maximum, and one to count the total number of packets dropped because the
queue was full. Then make them available as a read only value somewhere
in the proc/sys/net area.

-- 
Stuart Lynne <sl@fireplug.net>      604-461-7532      <http://edge.fireplug.net>
PGP Fingerprint: 28 E2 A0 15 99 62 9A 00  88 EC A3 EE 2D 1C 15 68

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/