[PATCH 0/6] files: scalable fd management (V4)

[Dipankar Sarma]

This is an updated version of patchset published earlier
at - http://marc.theaimsgroup.com/?l=linux-kernel&m=111747394710704&w=2

Since then I have done the following additional testing :

1. SMP/UP kernels with and without CONFIG_PREEMPT on x86 and ppc64.
   running various tests like ltp, dbench, tiobench, reaim. 

2. Testing with a testcase that creates > 1024 fds and exercises
   the vmalloc allocation and freeing path.

3. More 24+ hour runs on both x86 and ppc64

4. Touch testing with X running on a desktop.

5. Testing with __ARCH_HAS_CMPXCHG undefined. I booted and ran some
   basic tests with this on a ppc64 SMP box in order to exercise
   the hashed locking.

Additional performance #s :
---------------------------

tiobench on a 4-way ppc64 system :
                                        (lockfree)
Test            2.6.10-vanilla  Stdev   2.6.10-fd       Stdev
-------------------------------------------------------------
Seqread         1428            32.47   1475.0          29.11
Randread        1469.2          17.27   1599.6          35.95
Seqwrite        262.06          9.31    246.8           30.94
Randwrite       548.38          12.49   521.4           61.98

With LL/SC based locks, cache line bouncing effect of file_lock
is not as pronounced, but it still makes a difference 
with seq and random reads.

Andrew, would it be possible to give this some testing time
in -mm ? If so, please let me know what would be an appropriate
time for that and I will send patches against -mm.

Thanks
Dipankar
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/