On Fri, Aug 21, 2015 at 02:56:48PM -0700, Andy Zhou wrote: > On Fri, Aug 21, 2015 at 10:48 AM, Ben Pfaff <b...@nicira.com> wrote: > > On Tue, Aug 11, 2015 at 05:55:14PM -0700, Andy Zhou wrote: > >> Json string are usually created and freed immediately. Thus, there > >> should not be any downside in creating a larger buffer initially to > >> avoid the cost of moving strings around in memory due to realloc() > >> call. > >> > >> The following script is used as benchmark to measure number of > >> bytes reallocated: > >> > >> ovs-vsctl add-br br0 > >> for i in `seq 0 100`; do > >> ovs-vsctl add-port br0 p$i > >> done > >> ovs-vsctl del-br br0 > >> > >> 'ds_bytes_relocated' coverage counter shows that the bytes relocated > >> are 2.5Mbytes/sec and 17Kbytes/sec, before and after this patch applied, > >> respectively. > > > > Is there a noticeable performance difference? > > Unfortunately, No. Using the script mentioned in the commit message, > but increase the number > of ports to 500, I noticed on average, there are about 3% drop (0.1ms > drop over 3.3s) in user space time. > It is not very noticeable. It is also within the variants of the tests.
I'm really uncertain about patches 7 and 8 because they didn't make a significant difference. I'm happy with the rest of the patches. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
