Re: Recent MAC improvements: Test Results

I know everyone is probably sick of these test results but I only have
one last set to share. The executive summary is that the mac context
recycling really does provide a notable performance boost. The use of
umac also provides a boost but not nearly as dramatic.

As a note: The client was using OpenSSL 0.9.7e. The server was using
0.9.7a. I also ran tests from and against OpenSSH 3.8.1p1 with no
problems except for umac failing.

The first line is the test type, average throughput and percent increase
in throughput performance compared to STD. The next line is the average
time and the standard deviation. The 3rd line is user time and SD, the
last is system time and SD. As a note, I'm not comfortable with the
results from the /dev/zero tests. There may have been contention on the
line causing problems. The large SD may support that. I'm rerunning the
tests on a cleaner path. If anyone wants those results just send me some
mail.

STD /dev/zero 21.3MB/s 0%+
96.11 19.14251411
37.4725 0.205653643
7.5275 0.207969507

MAC /dev/zero 44.9MB/s 110.9%+
45.55 0.403944694
34.3865 0.11586176
8.612 0.268241839

UMAC /dev/zero 46.3MB/s 117.1%+
44.28 2.828852473
32.829 2.936800801
8.765 0.211149735

HPN /dev/zero 22.9MB/s 7.3%+
89.59 19.79009276
37.468 2.853274389
7.3665 0.456926517


STD scp 30.1MB/s 0%+
68.06 1.233210935
0.4975 0.032097713
3.999 0.10789371

MAC scp 34.0MB/s 14.0%+
60.20 4.034492435
0.5455 0.067782547
4.242 0.403453513

UMAC scp 34.6MB/s 15.8%+
59.26 3.434033381
0.528 0.055212508
4.1835 0.324982186

HPN scp 30.4MB/s 0.9%+
67.42 1.553637159
0.1445 0.019594575
3.2015 0.087676139
_______________________________________________
openssh-unix-dev mailing list
openssh-unix-dev@mindrot.org
http://lists.mindrot.org/mailman/lis...enssh-unix-dev