[OpenAFS-devel] read performance 1.5.74 Linux

[Simon Wilkinson]

On 14 May 2010, at 10:25, Hartmut Reuter wrote:

> /afs/.notebook/u/hwr: read_test 100mb
> open of 100mb bytes took 0.000 sec.
> read of 100000000 bytes took 327.955 sec.
> Total data rate =3D 298 Kbytes/sec. for read
> /afs/.notebook/u/hwr:
>=20
> Then I applied patch I1a11798f to rx.c and built 1.5.74-osd again.
> There is a little improvement, but it's still very slow:
>=20
> ~/a/n: read_test 100mb
> open of 100mb bytes took 0.067 sec.
> read of 100000000 bytes took 29.154 sec.
> Total data rate =3D 3342 Kbytes/sec. for read
> ~/a/n:

Maybe I'm misunderstanding you here, but the data rate above looks like =
it's more than 10 times faster than that in your earlier test, which =
seems like more than a little improvement. 3Mbytes/sec still seems slow, =
though - the tests I've done show data rates (with 1.5) of the order of =
40Mbytes/sec. What figure do you see if you test using an unpatched =
1.4.x cache manager?

> I think before giving this source to the users as 1.6 this problem has =
to be
> understood and solved!

If you can reproduce this using a command line tool such as afsio, then =
we could trying and track this down using git bisect - that's probably =
going to be the best route. Testing with afsio would also let us =
implicate the RX library, rather than the cache manager, as being the =
source of the problem.

Cheers,

Simon.