http://vincent.bernat.im/en/blog/2011-ssl-benchmark.html Interesting test results comparing and analyzing SSL performance. Using right combination of software and proper tunning, you can get up to 14000 TPS (re-shake every 80 requests) on HP DL 380 G7, with two Xeon L5630 (running at 2.13GHz for a total of 8 cores), without hyperthreading, using a 2.6.39 kernel (HZ is set to 250) and two Intel 82576 NIC.
http://vincent.bernat.im/en/blog/2011-ssl-benchmark.html Interesting test results comparing and analyzing SSL performance. Using right combination of software and proper tunning, you can get up to 14000 TPS (re-shake every 80 requests) on HP DL 380 G7, with two Xeon L5630 (running at 2.13GHz for a total of 8 cores), without hyperthreading, using a 2.6.39 kernel (HZ is set to 250) and two Intel 82576 NIC.
http://blog.exceliance.fr/2011/09/16/benchmarking_ssl_performance/ SSL perfomance benchmark on Atom D510 ( dual core 1.6 GHz ). (STUD/OpenSSL 0.9.8) . With re-negotiation every 100 requests and object size = 4K, 2300 SSL Req/Sec on Intel Atom!
http://blog.exceliance.fr/2011/09/16/benchmarking_ssl_performance/ SSL perfomance benchmark on Atom D510 ( dual core 1.6 GHz ). (STUD/OpenSSL 0.9.8) . With re-negotiation every 100 requests and object size = 4K, 2300 SSL Req/Sec on Intel Atom!
https://community.openvpn.net/openvpn/wiki/Gigabit_Networks_Linux
http://haproxy.1wt.eu/10g.html
http://haproxy.1wt.eu/10g.html HA Proxy [synthetic] benchmarks on 10Gbit traffic. Linux network stack is the bottle neck here, not HAProxy, i.e. you can 10G line rate on big frames and few connections.
http://vbtechsupport.com/606/ Sysbench-based benchmark comparing OLTP workloads for different builds and forks of MySQL on 64-core intel machine (Quad Intel Xeon L7555 ) . Only one definitive result is that MySQL cannot scale beyond 8 or 16 cores, or in some cases beyond just 2 cores (booo! ). In all cases performance went down when more than 24 cores were used .
http://www.facebook.com/notes/facebook-engineering/real-world-web-application-benchmarking/203367363919 Facebook has build custom web-app testing framework to test server performance. Read it, especially the section on why “standard” SPECxxx benchmarks are not a good measure of anything. Also, ready about performance-per-watt, as this is really important for modern data centers.
http://www.bureausolomatina.ru/sites/default/files/SPICE%20Benchmark%20-%202010-11-16.pdf Benchmark comparing SPICE with RDP. In short, network utilization is similar, SPICE works for with video, but requires ~ 8Mbps of bandwidth. (SPICE work only for VMs)