This is the mail archive of the
ecos-patches@sources.redhat.com
mailing list for the eCos project.
Re: gradual increase in ip_nfragpackets
Hi, Jifl;
On Mon, Jun 16, 2003 at 05:37:11AM +0100, Jonathan Larmour wrote:
> Motoya Kurotsu wrote:
> > Hi, all,
> >
> > The attachment is the patch to fix the problems below;
> >
> > 1) ip_nfragpackets is increasing by dropping fragments due to mbuf
> > starvation and eventually exceeds ip_maxfragpackets, which
> > results in refusing all fragments later.
> > 2) maxfragpackets and maxnipq functions implemented approximately
> > the same limits on fragment memory usage, but in different fashions.
>
> I would like to apply this *but* I would ask you if you can let me know
> what kind of testing you have done on it first?
This bug was found during excuting my endurance test program and my patch
was verified by the same program. This program generates a lot of maximum
length of UDP packets(ip length = 65535 byte) and send them to the destination
(in this case, my box) such at high speed that my box frequently runs short
of mbuf. Then my box is configured as a simple UDP packet receptor. Although
this program can send any number of packets, I tested 19200 packets at most.
Motoya Kurotsu
Allied Telesis K.K.