UDP queue depth

Michael O'Donnell mod+gnhlug at std.com
Thu Jun 19 16:40:16 EDT 2003


Given a "normal" Linux box of some recent vintage
(like, say, 2.4.18) can anybody help me get a sense
of how many UDP datagrams of some nominal size (like,
say, 1k or 4k) can be received before the kernel
(as it is entitled to do with UDP) starts dropping
them on the floor?  In other words, assuming I've
rigged up the sender and the receiver properly but
then had the receiver intentionally fail to read the
incoming datagrams, how soon would the kernel start
to discard new arrivals?

I'm not necessarily looking for hard numbers, just to
get the general idea, so keep the scenario as simple
as you like; for example, we can assume that the UDP
traffic in question is the only traffic on the wire,
and if some parameter (like, say, transmission rate)
matters, just pick a value for that parameter.

I'd think it'd be fairly straightforward to write
a test program to determine this empirically (and
rather less straightforward but still feasible to
RTFSC for the Linux kernel) but if somebody happens
to have the numbers (or a pointer to the numbers)
handy that'd be great.




More information about the gnhlug-discuss mailing list