On Fri, May 31, 2013 at 01:33:46PM -0400, Jeff Morriss wrote:
> On 05/31/13 12:59, Jakub Zawadzki wrote:
> > Has anyone performed some benchmarks before/after this patch? :)
>
> It appears to have a fairly negligible effect. Here's a comparison
> using tools/test-captures.sh (on a pool of files):
> [...]
>
> (That's about a 0.46% increase.)
>
> And here's one focusing just on the "no tree" case:
>
> [...]
>
> (That's about a 1.36% increase.)
my tests shows about 4% (compiled with -O1)
ten times:
$ tshark -r /tmp/b.pcap -n > /dev/null
r49643 avg: 1m40.054s
r49652 avg: 1m44.134s
$ capinfos /tmp/b.pcap
Number of packets: 7284 k
File size: 944 MB
perf tool shows ~ 2% total overhead
$ perf record tshark -r /tmp/b.pcap -n > /dev/null
$ perf report
11.69% tshark libc-2.16.so [.] vfprintf
10.97% tshark libc-2.16.so [.] 0x000000000007de90
5.09% tshark libc-2.16.so [.] _IO_default_xsputn
2.41% tshark libwireshark.so.0.0.0 [.] ieee80211_radiotap_iterator_next
...
1.63% tshark libwireshark.so.0.0.0 [.] get_hfi_and_length
...
0.39% tshark libwireshark.so.0.0.0 [.] test_length
Ok, let's leave it :)
Cheers,
Jakub.
