Messages in this thread | | | Date | Thu, 10 Jan 2019 14:51:53 +0100 | From | Greg Kroah-Hartman <> | Subject | Re: [PATCH v1 0/3] reduce tty latency |
| |
On Thu, Jan 10, 2019 at 11:12:29AM +0100, Oleksij Rempel wrote: > This patch set is reducing latency on tty path. > For testing I used hackbench running on all cores of 4 core system and > high prioritized application sending and receiving packets over tty interface > with loop-back adapter.
Odd, all of these ended up in my spam folder, you might want to check your server email settings...
> Results without this patches: > latency histogram: > 0 ... < 250 usec : 1933104 transmissions > 250 ... < 500 usec : 21339 transmissions > 500 ... < 750 usec : 8952 transmissions > 750 ... < 1000 usec : 6226 transmissions > 1000 ... < 1500 usec : 7688 transmissions > 1500 ... < 2000 usec : 5236 transmissions > 2000 ... < 5000 usec : 11724 transmissions > 5000 ... < 10000 usec : 3588 transmissions > 10000 ... < 50000 usec : 2123 transmissions > 50000 ... < 1000000 usec : 20 transmissions > >= 1000000 usec : 0 transmissions > > Test results after this patches: > min latency: 0 sec : 75 usec > max latency: 0 sec : 125 usec > average latency: 81 usec > latency measure cycles overall: 79000000 > latency histogram: > 0 ... < 250 usec : 79000000 transmissions > 250 ... < 500 usec : 0 transmissions > 500 ... < 750 usec : 0 transmissions > 750 ... < 1000 usec : 0 transmissions > 1000 ... < 1500 usec : 0 transmissions > 1500 ... < 2000 usec : 0 transmissions > 2000 ... < 5000 usec : 0 transmissions > 5000 ... < 10000 usec : 0 transmissions > 10000 ... < 50000 usec : 0 transmissions > 50000 ... < 1000000 usec : 0 transmissions > >= 1000000 usec : 0 transmissions > average no. of read calls to assemble the packet: 1
Like Linus said, who runs a real-world system that cares about this latency measurement?
Yes, it might be fun for odd benchmarks to show the value of one RT patchset/OS vs. another one, but this change can cause real issues in real systems that do real, non-serial-loopback work.
thanks,
greg k-h
| |