[OpenAFS] Network becomes terribly slow when cache manager flushes updates over xDSL

fredericg_99@yahoo.fr fredericg_99@yahoo.fr
Wed, 8 Jul 2009 08:19:15 +0200 (CEST)


> >> My feeling is that #define RX_MAX_FRAG 1 instead of 4 does make
> >> things
> >> behave better if there is packet loss to be expected.

I tried this, and i does really change everything... (mea culpa, there seem=
 to be packet losses on my line...)
I still come to have ping ~500ms sometimes, but it does not seem to be rela=
ted with my activity on the AFS filesystem. Merely a random connection prob=
lem, as it may "normaly" happen.

So, say the problem is mostly solved. Firefox even works quite flawlessly..=
.

Still, I tried the "ping -M do -s xxx yyy", and I came to something strange=
 :=20
s=3D1500 -> error "icmp_seq=3D1 Frag needed and DF set (mtu =3D 1500)"
s=3D1400 -> ping OK (104ms avg) :
   "PING addr.net (x.x.x.x) 1400(1428) bytes of data", then,
   "1408 bytes from addr.net (x.x.x.x): icmp_seq=3D1 ttl=3D47 time=3D103 ms=
"
I don't really understand why there are 3 different numbers : 1400, 1428, a=
nd 1408.
Then I tried to find the "limit", but after few attemps (over 1470, where i=
t seemed to be), the error message changed to=20
   "icmp_seq=3D1 Frag needed and DF set (mtu =3D 1492)"

Why this drop from 1500 to 1492? It suddenly met M. Colomb? :-)

Now it does not seem to drop again, and the limit before fragmentation occu=
rs is definitely "-s 1466" (1492 minus 28).

I have to say that ifconfig still shows a MTU of 1500.

This does not seem to be a real issue, but maybe unerstanding this will hel=
p me tuning the parameters...

By the way, I didn't get the final point about the "-nojumbo" and "-rxmaxmt=
u" params : will they appear in a release soon? And if yes, they will repla=
ce the change to RX_MACK_FRAGS I had to make?

Thanks again to your help, and, I can say it now without hesitations : than=
ks for your great filesystem, as it really has good performances even on st=
andard xDSL links :-)

Frederic Grelot.

> It could also be the MTU over DSL is smaller then you think, and so
> many/most packets
> are getting fragmented.  A mod is being added in 1.4.11 to add
> -rxmaxmtu N to afsd
> that can work on MAC or any Unix system. Set this to 56 bytes less
> then the actual
> MTU being used. Use ping  with the don't fragment options to test at
> what size fragmentation
> occurs.
>   ping -M do -s nnnn ...
>=20
> The Windows version already has a registry setting to do the same
> thing.
>=20
>=20
>=20
>=20
> > Thanks again for your help.
> >=20
> > Fr=C3=A9d=C3=A9ric Grelot.
> > _______________________________________________
> > OpenAFS-info mailing list
> > OpenAFS-info@openafs.org
> > https://lists.openafs.org/mailman/listinfo/openafs-info
> >=20
> >=20
>=20
> --=20
>=20
>   Douglas E. Engert  <DEEngert@anl.gov>
>   Argonne National Laboratory
>   9700 South Cass Avenue
>   Argonne, Illinois  60439
>   (630) 252-5444