Re: 2nd round of xen benchmarking

[mlelstv%serpens.de@localhost (Michael van Elst)]

bouyer%antioche.eu.org@localhost (Manuel Bouyer) writes:

>PV (netbsd-XEN3_DOMU.gz from the NetBSD 10.0 release)
>      155.66 real        64.30 user        91.96 sys
>      153.36 real        63.57 user        92.71 sys
>      153.79 real        64.68 user        92.20 sys
>      153.77 real        64.89 user        91.99 sys

>PVH (netbsd-GENERIC.gz from the NetBSD 10.0 release)
>       64.08 real        38.36 user        19.98 sys
>       61.28 real        38.80 user        20.05 sys
>       61.31 real        38.79 user        20.07 sys
>       61.38 real        39.08 user        19.83 sys

>As expected, with EPT PVH gives a good performance boost over PV.
>I can't test bare metal as this is a production system running several VMs.


I've tested just the 'make' (no 'make clean') and SSD instead
of tmpfs) on a i5-2520M (dual core + HT) running 10.99.11.

As a comparison, some openssl speed benchmarks for something that should
be less affected by hypervisor calls and MMU tables.


DomU (one vpcu, cpus="2", i.e. binding to the second physical core):

82.564u 120.134s 3:29.73 96.6%  857+416k 0+20101io 43pf+0w
83.227u 119.156s 3:30.40 96.1%  858+411k 74+20126io 1774pf+0w

AES-128-CBC 70863.43k  80367.31k  83191.41k  84202.06k  84429.31k  84494.63k
ChaCha20   206227.66k 388520.22k 824340.16k 893265.14k 912746.02k 912681.33k


against Dom0 (dom0_max_vcpus=1, no guest)

60.420u 107.238s 2:54.59 96.0%  1036+500k 122+1393io 1835pf+0w
61.254u 107.165s 2:55.10 96.1%  1032+498k 119+1378io 1835pf+0w

AES-128-CBC 68896.51k  79475.74k  82082.92k  83084.77k  83316.83k  83319.50k
ChaCha20   203695.15k 384909.19k 813382.84k 881179.87k 901193.83k 901486.15k


against Bare metal: (all CPUs, or just just cpu#0 online, no difference):

44.286u 21.338s 1:08.30 96.0%   2649+1105k 112+1327io 1922pf+0w
44.048u 20.948s 1:10.61 92.0%   2674+1268k 117+928io 1836pf+0w

AES-128-CBC 82885.72k  95674.52k  99363.95k  100529.43k  100911.79k  100886.35k
ChaCha20   247237.43k 466793.66k 989934.51k 1066972.84k 1091012.80k 1093120.17k


Disabling CPU turbo mode by setting machdep.cpu.frequency.target=2500
(instead of 2501) gets openssl numbers close to the Dom0 case, but
still a huge difference for the m4 build.

51.214u 25.220s 1:18.23 97.6%   2274+957k 5+1318io 74pf+0w
51.467u 25.040s 1:20.42 95.1%   2272+1158k 120+1307io 1945pf+0w

AES-128-CBC 69551.28k  80083.43k  82532.44k  83645.21k  84365.58k  84369.13k
ChaCha20   227950.31k 431182.74k 830283.30k 888214.05k 905877.97k 908288.34k



Greetings,