[zeromq-dev] Windows allocator performance under high load

Steven McCoy steven.mccoy at miru.hk
Tue Mar 29 15:17:54 CEST 2011

This was interesting, a quick comparison of Microsoft's CRT version 10,
Nedmalloc from git head, and Microsoft's Concurrency Runtime (C++).

At high but stable PGM load of 50,000 packets-per-second at 414mb/s wire
speed on gigabit fibre, Microsoft's CRT concludes with a latency of ~5ms,
Nedmalloc reaches ~3.3ms, and Microsoft's Concurrency CRT reaches ~3ms.
 Linux and ptmalloc3 reach 83µs for comparison.

Unfortunately jemalloc and tcmalloc aren't being conducive for testing but
presumably sit near Nedmalloc's performance, on Linux both have tested worse
than ptmalloc3.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20110329/69afe04a/attachment.htm>

More information about the zeromq-dev mailing list