> I don't yet have any useful data on speed improvements from both changes.
> The plan is to to collect them and report back in the next few days.
> If we don't want to apply them without the performance data, then
> please consider this posting as RFC.
After spending some time on performance measurements, and some
discussion with Ben, I figured out a way to
measure performance improvements implemented by this patch series:
The main idea is to measure the timing of
multiple back to back transactions.
Specifically, I use :
ovs-vsct add-port br0 p1 -- add-port br0 p2 -- ... add-port br0 p20
to add 20 ports in one go, and measure the delay in executing the
entire command. This would show ovsdb-server's work
load of not only handling each transaction, but also the background
work required to finish servicing the transaction.
Here are the results:
Without this patch series:
==================
#monitors delay of 20 transactions
1 0.046s
100 0.532s
200 1.007s
300 1.599s
With this patch series:
=================
#monitors delay of 20 transactions
1 0.046s
100 0.047s
200 0.047s
300 0.049s
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
