Dummynet low bandwidth simulation

Luigi Rizzo rizzo at icir.org
Mon Apr 26 15:39:26 PDT 2004

are you sure you aren't running out of mbufs ?
netstat -m should tell you.

Additionally, note that the 250ms of delay are probably way too much
for your config (you are adding 250ms each way, which makes a 500ms RTT,
not accounting for transmission times -- i think normal values here
are more like 150-200ms rtt), and possibly irrelevant given how
large your queues are -- a full-size packet is 12000 bits or 200ms,
you can have up to 20 queued...


On Mon, Apr 26, 2004 at 11:58:21PM +0200, Ivan Voras wrote:
> I'm trying to see how a web server performes when "attacked" by many low 
> bandwith connections, and my idea is to use dummynet to simulate such 
> conditions from a single client computer on the same (100Mbps) LAN. I've 
> created a pipe:
> ipfw pipe 1 config bw 56Kbit/s delay 250 queue 20 buckets 1024 mask all
> My rules are:
> ipfw pipe 1 tcp from me to webserver dst-port 80
> ipfw pipe 1 tcp from webserver 80 to me
> When I start the benchmark program (siege) with relatively low 
> concurrency (50 clients), everything works as expected, but after 
> increasing to 100+ simultaneous requests, I start getting huge amounts 
> of "read error: Connection reset by peer" and "broken pipe" messages. 
> During that time the load on the web server is low (plenty of idle time 
> and memory).
> Without traffic shaping, the server handles 250-300 concurrent requests 
> without problems in the same setup.
> I've tried changing the queue and bucket size (and adding 'noerror'), 
> with no results. Eliminating latency (delay 0) "fixes" the problem, but 
> it's not meaningfull that way.
> Am I missing something?
> -- 
> C isn't that hard: void (*(*f[])())() defines f as an array of
> unspecified size, of pointers to functions that return pointers to
> functions that return void.
> _______________________________________________
> freebsd-current at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe at freebsd.org"

More information about the freebsd-current mailing list