Re: FreeBSD performance on single CPU.

From: Ivan Voras <ivoras_at_freebsd.org>
Date: Wed, 19 Nov 2008 16:39:48 +0100
Channa wrote:
> Hi,
> Thank you very much for your information.
> 
> I am using current implementation of FreeBSD malloc.
> I am checking the performance on single CPU machine.
> 
> I am using it in my own environment i am not using FreeBSD7.0.
> 
> I disabled the MALLOC_DEBUG option in the malloc implementaton
> and checked the performance i see no difference the results are the same.
> 
> ./mallco-test
> Starting test...
> Thread -1101005488 adjusted timing: 4.495931 seconds for 1000000
> requests of 512 bytes.
> 
> Could please tell me if anything else needs to be changed.?

Yes, you need to disable kernel debugging also. (you have recompiled
libc after removing MALLOC_DEBUG, right?).

Also, it looks like your comparisons have problems. You are changing the
number of requests and request size.

Finally, to what are you actually comparing the malloc performance to?
Why are you trying to benchmark it?


> 2008/11/19 Ivan Voras <ivoras_at_freebsd.org>:
>> Channa wrote:
>>> Hi,
>>>
>>> I am using FreeBSD malloc from the current working branch.
>>> I hope its jemalloc.
>>>
>>> I tested it on a single CPU machine i could get the following results
>>>
>>> # ./malloc-test  1024 10000000 4
>>>
>>> Starting test...
>>> Thread -1096811184 adjusted timing: 102.369100 seconds for 10000000
>>> requests of 1024 bytes.
>>> Thread -1101005488 adjusted timing: 103.212512 seconds for 10000000
>>> requests of 1024 bytes.
>>> Thread -1098908336 adjusted timing: 103.491399 seconds for 10000000
>>> requests of 1024 bytes.
>>> Thread -1094714032 adjusted timing: 103.605124 seconds for 10000000
>>> requests of 1024 bytes.
>>>
>>> I checked the result in the FreeBSD mailing list link given below:
>>>
>>> http://unix.derkeiler.com/Mailing-Lists/FreeBSD/current/2005-12/msg00294.html
>>>
>>> The jemalloc gives very good results. Is that the performance is good
>>> only on SMP?
>>>
>>> Or on single processor also it should perform well?
>>>
>>> But on single CPU i could see bad results.
>>>
>>> Could anyone help me out?
>> Your message is not very clear but here are some things that might help you:
>>
>> 1) -CURRENT has debugging enabled both in kernel and in malloc. You need
>> to disable both before benchmarking anything.
>> 2) According to the post you linked, jemalloc should be 1.1 times faster
>> for single-threaded processes than phkmalloc, on that particular
>> benchmark. This benefit will probably also be visible on single-CPU
>> machines.
>> 3) You don't need to run -CURRENT to get jemalloc - it is also prosent
>> in 7.0.
>>
>>
> _______________________________________________
> freebsd-current_at_freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
> 


Received on Wed Nov 19 2008 - 14:39:08 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:37 UTC