Solaris fcntl bug 4700402

Bogdan Iamandei bogdan at its.uq.edu.au
Fri Oct 11 06:30:00 GMT 2002


Tristan Ball wrote:
> I've also just recieved this t-patch, and I'll be testing this weekend. 
> I'll post results asap.

We've received the patch quite some time ago - and so far I can say that
it seems a good and *freakin' fast* one. <digging for the results> ah -
there they are.

> 
> According to my contact at sun, they are looking at a Jan/03 date for 
> the official release.

[...this is a snippage from an internal mail...]

     I have finished the fcntl-tests on Unlucky. As you might
have heard  me before - they are quite impressive. The locker program
goes amazingly faster and even if the machine is loaded, it is still
quite responsive. As a matter of fact - terminals no longer seem locked
when under load.

     The samba torture test also shows radical improvement. I could
go as high as 250 simultaneous processes before, because of some
limitations occuring also on BSD (Chris Smith confirmed this) hit and
the tests were aborted.

     Here are some results - to judge for yourself:

Locker
------

800 processes - 50 locks

        BEFORE                              AFTER

     12mins 55sec                        00mins 07sec

900 processes - 50 locks

     16mins 57sec                        00mins 10sec

1000 processes - 50 locks

     26mins 25sec                        00mins 13sec

1100 processes - 50 locks

     36mins 50sec                        00mins 15sec

1200 processes - 50 locks

     59mins 40sec                        00mins 21sec

     At this point I no loger have data for the "BEFORE" case, so all
I can  say is that I went as high as 4100 processes with 50 locks and I
got 6mins and 50 seconds to completion.

     The most interesting thing to note is that by far the most
time-comsuming task is forking the processes, and that once they're
created, draining the locks in the queue only takes a few seconds, where
as before it would've taken a comparable amount of time to the creation
time.

Samba Torture
-------------
            BEFORE                   AFTER
100 clients    1673.03 secs           750.573 secs
                78.8988 MBit/sec       175.866 MBit/sec

150 clients    3159.51 secs           1213.71 secs
                62.668 MBit/sec        163.136 MBit/sec

200 clients    NO DATA                1905.17 secs
                                       138.571 MBit/sec

250 clients    NO DATA                2686.8 secs
                                       122.823 MBit/sec

[...end of snippage...]


	I hope you find these results interesting. Unfortunately I
cannot say anything about how this patch behaves in productions since
it was decided that it shall not be inflicted upon our users just yet.

	Questions? Anyone? Drool? :)

						Ino!~

-- 
I have seen things you people wouldn't believe.  Attack ships on fire
off the shoulder of Orion.  I watched C-beams glitter in the dark
near the Tannhauser Gate.  All those moments will be lost in time,
like tears in rain.  Time to die.




More information about the samba-technical mailing list