Hi Denys,
The report shows OpenSSL library doing small SHM allocations (4, 10,
608, 24... bytes), which seem to frequently take place inside the
PEM_read_bio_X509() loop (as part of the load_certificate_db()
function). Such a sequence of allocations could be stress-testing the
allocator in a way that could justify 250 ms per certificate in total,
as it is fragmenting the memory. The effect can be more pronounced the
*less* stuff is going on in your OpenSIPS instance, as the process of
breaking up the big memory chunk into smaller units may use up to
hundreds of cycles on each allocation. For example: testing box with
no SIP traffic, or 'tls_reload' after a fresh restart, etc.
Please try the following:
- still using F_MALLOC, try doing more 'tls_reload' operations in a
row. Does performance improve?
- try using the "-a HP_MALLOC" allocator when booting your OpenSIPS -
that one favors memory fragmentation a bit more, so subsequent reloads
should be faster
Best regards,
Liviu Chircu
www.twitter.com/liviuchircu | www.opensips-solutions.com
On 14.05.2024 16:47, Denys Pozniak wrote:
I disabled logging and added some resources to the virtual machine.
On a working OpenSIPS, I reloaded the tls several times and in
parallel ran a trap.
#opensips-cli -x mi tls_reload
#opensips-cli -x trap
If possible, please analyze it again, maybe you could find something
interesting:
https://github.com/denyspozniak/opensips_tls_debug/tree/main
Thanks in advance!
_______________________________________________
Users mailing list
[email protected]
http://lists.opensips.org/cgi-bin/mailman/listinfo/users
