[Jool-list] NAT64 performance

[Sander Steffann]

Hi,

The great Cisco TRex team [1] has updated their performance testing software so that it can now test NAT64 boxes! They added an Advance Stateful mode (ASTF) that implements an extremely fast TCP stack for testing [2]. Naturally I pointed it at a Jool box :-)  Unfortunately the Jool box collapsed under the surge of new sessions :-(

Some observations:
- At the beginning of the test the performance seems fine (2.5Gbit/s in my test)
- When a lot of flows are created the box starts dropping packets massively and performance collapses
- The BIB logging actually still shows new sessions being created

From this it seems that the session table is a performance bottleneck. Not surprising for a stateful protocol :-)  But because Jool still creates new sessions but starts dropping packets of existing sessions it also seems that session creation has a higher priority than packet forwarding. Could it be that session creation locks the session table, and that the increased size of the table keeps the locks blocked for longer and longer?

Q1: any disagreements with my interpretation? ;-)
Q2: what can we do to improve this?

Cheers,
Sander

PS: Developers that want to work on this and would like access to my lab boxes (Dell R630 with lots of cores, memory and some 10Gbit/s NICs): feel free to contact me!

[1] https://trex-tgn.cisco.com
[2] https://trex-tgn.cisco.com/trex/doc/trex_astf.html

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 455 bytes
Desc: Message signed with OpenPGP
URL: <http://mail-lists.nic.mx/pipermail/jool-list/attachments/20170904/fd84458b/attachment.bin>