Lockless uidinfo.

[Pawel Jakub Dawidek]

On Sun, Aug 19, 2007 at 01:09:17AM +0200, Pawel Jakub Dawidek wrote:
> Ok, after implementing atomic_fetchadd_long() on amd64, we get additional
> 6% of performance improvement:
> 
> x ./uidinfo_lockfree.txt (atomic_cmpset_long loop)
> + ./uidinfo_waitfree.txt (atomic_fetchadd_long)
> +------------------------------------------------------------------------------+
> |                                                                             +|
> |                                                                             +|
> |x   xx    xx                                                              + ++|
> |  |__MA___|                                                                |AM|
> +------------------------------------------------------------------------------+
>     N           Min           Max        Median           Avg        Stddev
> x   5       1561566       1575987       1568964       1569767     5853.1399
> +   5       1662362       1665936       1665810     1664881.8     1541.2693
> Difference at 95.0% confidence
>         95114.8 +/- 6241.96
>         6.05917% +/- 0.397636%
>         (Student's t, pooled s = 4279.88)

One more thing - comparsion between waitfree method and when chgsbsize()
is a no-op:

x ./uidinfo_waitfree.txt (atomic_fetchadd_long)
+ ./uidinfo_none.txt (no chgsbsize)
+------------------------------------------------------------------------------+
|x xx                                                +  +    +   +            +|
| |AM|                                               |_______M_A________|      |
+------------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x   5       1662362       1665936       1665810     1664881.8     1541.2693
+   5       1718287       1744448       1726343     1728176.6     10271.936
Difference at 95.0% confidence
        63294.8 +/- 10711.8
        3.80176% +/- 0.643395%
        (Student's t, pooled s = 7344.66)

This small speed up is of course because of atomics in waitfree case, but it
shows that we can't do much better than waitfree version, as ideal case is only
3.8% faster, at least for this benchmark.

This was for eight processes and this is for one:

x ./uidinfo_up_waitfree.txt (atomic_fetchadd_long)
+ ./uidinfo_up_none.txt (no chgsbsize)
+------------------------------------------------------------------------------+
|x           x   x          x                             x            +  + +++|
| |______________M_____A_____________________|                           |_AM_||
+------------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x   5        419799        431017        422877      424180.2     4265.4168
+   5        433705        434955        434696      434509.6        519.68
Difference at 95.0% confidence
        10329.4 +/- 4431.34
        2.43514% +/- 1.04468%
        (Student's t, pooled s = 3038.41)

This means that something else slows that eight processes case a bit, but it's
not uidinfo.

Ok, I need to stop, I need to stop right now!

-- 
Pawel Jakub Dawidek                       http://www.wheel.pl
pjd at FreeBSD.org                           http://www.FreeBSD.org
FreeBSD committer                         Am I Evil? Yes, I Am!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-arch/attachments/20070819/adc29f82/attachment.pgp