mbox series

[0/2] Reusing nfct handle for bulk ct loads

Message ID 20210823155715.81729-1-mikhail.sennikovskii@ionos.com
Headers show
Series Reusing nfct handle for bulk ct loads | expand

Message

Mikhail Sennikovsky Aug. 23, 2021, 3:57 p.m. UTC 
Hi Pablo & all,

I've finally got my hands on doing some more sophisticated testing
of the bulk ct operations functionality (-R option), and revealed 
a way to makethe ct load operations to be almost twice as fast 
by reusing the same nfct handleinstead of recreating it for each
entry.
This becomes essential for bulk loads of big amounts of entries,
where the overall speedup reaches values of seconds.

The two patches posted here are the tests/conntrack/bulk-load-stress.sh 
script that could be used for stress testing the bulk operations
and the patch improving the bulk ct load performance.

(Btw the script also reveals that bulk deletion of big amount
of entries specified individually is really slow,
but it's another story which could be addressed later if needed.)

Any feedback/suggestions for the patches are welcome.

Regards,
Mikhail


Mikhail Sennikovsky (2):
  tests/conntrack: script for stress-testing ct load
  conntrack: use same nfct handle for all entries

 src/conntrack.c                     | 160 +++++++++++++++++-----------
 tests/conntrack/bulk-load-stress.sh | 152 ++++++++++++++++++++++++++
 2 files changed, 247 insertions(+), 65 deletions(-)
 create mode 100755 tests/conntrack/bulk-load-stress.sh