New malloc ready, take 42

David Xu davidxu at freebsd.org
Wed Dec 28 21:05:58 PST 2005


Jason Evans wrote:
> 
> On Dec 28, 2005, at 8:52 PM, David Xu wrote:
> 
>> Jason Evans wrote:
>>
>>> So, how about it?  Is jemalloc ready to go in now?
>>> Thanks,
>>> Jason
>>
>>
>> I have tested super-smack with your patch on my Pentium-D machine:
>> CPU   : P4 dual-core 2.8Ghz
>> Memory: 1G bytes
>> I am using libthr.
>>
>> phkmalloc can reach about 21000 query-per-s.
>> jemalloc can only reach about 15000 query-per-s,
>> about 28% performance is dropped.
> 
> 
> That is a much different result than Kris Kennaway got on a dual-dual  
> Opteron system, as well as my results on a 3.2 GHz P4.  Can you  please 
> provide details on what version of the jemalloc patch you  used, which 
> MALLOC_OPTIONS flags you used, what parameters you  specified to 
> super-smack, etc.?
> 
> A Pentium-D should behave similarly to two single-core CPUs, right?
> 
> Thanks,
> Jason
> 
> 
> 
I am using patch file jemalloc_20051222c.diff. I don't use
MALLOC_OPTIONS and don't create link malloc.conf in /etc.
I have tested super-smack with following command:

%super-smack ./select-key.smack 10 5000
Query Barrel Report for client smacker1
connect: max=4ms  min=1ms avg= 2ms from 10 clients
Query_type      num_queries     max_time        min_time        q_per_s
select_index    100000  0       0       15895.76
%super-smack ./select-key.smack 10 5000
Query Barrel Report for client smacker1
connect: max=3ms  min=2ms avg= 2ms from 10 clients
Query_type      num_queries     max_time        min_time        q_per_s
select_index    100000  0       0       16029.86
%super-smack ./select-key.smack 10 5000
Query Barrel Report for client smacker1
connect: max=2ms  min=0ms avg= 1ms from 10 clients
Query_type      num_queries     max_time        min_time        q_per_s
select_index    100000  0       0       14732.03




More information about the freebsd-current mailing list