Malloc scaling/speedup work, request for testers/reviews

Dylan Reinhold dylan at ocnetworking.com
Wed Apr 21 19:01:10 PDT 2010


Venkatesh Srinivas wrote:
Hiya,

I've been working for some time on improving the libc malloc in
DragonFly, specifically improving its thread scaling and reducing the
number of mmap/munmap system calls it issues.
  
Venkatesh,
 I ran a quick test on one of my machines. I used the LD_PRELOAD
I ran with 10,000 calls 1-1000 blocks, it shaved ~21 seconds off the the 
time.
The system is a AMD Athlon(tm) XP 1800+ (1493.66-MHz) / with 512megs of 
ram).
I tried with your 50,000 and 1-1,512 block size, but the system ran out 
of memory/swap and the utility died.

Pre -
dylan at backup_a:~/malloc$ ./sh6bench
call count [1000]: 10000
min block size [1]: 1
max block size [1000]:
Total elapsed time: 52.00 (51.5938 CPU)
Post -
dylan at backup_a:~/malloc$ ./sh6bench
call count [1000]: 10000
min block size [1]:
max block size [1000]:
Total elapsed time: 31.00 (30.9297 CPU)
Regards,
Dylan




More information about the Kernel mailing list