Latency

• CPU Clock Cycle: Modern CPUs from 800 MHz to 5 GHz
• L1 cache: Intel Haswell, 4 cycles (http://www.7-cpu.com/cpu/Haswell.html)
• L2 cache: Intel Haswell, 12 cycles (http://www.7-cpu.com/cpu/Haswell.html)
• Branch misprediction: Haswell, mOp cache hit, 15 cycles (http://www.7-cpu.com/cpu/Haswell.html)
• Mutex lock: https://gist.github.com/jboner/2841832
• malloc (1 KB): glibc, 1 KB
• page copy: 4K, x86 sse2 movdqa, 3 clocks per 128 bits
• win mutex lock: MSDN Magazine, Sep 2006, 9801 cycles, without contention
• compress 1K: 1K with zippy, from https://gist.github.com/jboner/2841832 = 3 us
• Gigabit ethernet: Send 1K, from https://gist.github.com/jboner/2841832 = 10 us
• Context switch: This is HIGHLY variable, depending on the size of the working set, cache coherency, SMP, TLB, etc. Taken from a graph at http://blog.tsunanet.net/2010/11/how-long-does-it-take-to-make-context.html
• SSD access: 1 page, approx 1 GB/s SSD
• Read 1MB from memory: sequentially
• Data centre network: roundtrip
• Read 1MB from SSD: sequentially
• Disk seek (10,000 RPM): Up to one full platter rotation, head seek time within range from https://en.wikipedia.org/wiki/Hard_disk_drive_performance_characteristics
• Disk seek (7,200 RPM): Up to one full platter rotation, head seek time within range from https://en.wikipedia.org/wiki/Hard_disk_drive_performance_characteristics
• Disk seek (5,400 RPM): Up to one full platter rotation, head seek time within range from https://en.wikipedia.org/wiki/Hard_disk_drive_performance_characteristics
• Network Latency: Various ping times usually in this range
• DNS latency: Cold, approx 3 network roundtrips

As Operations per Second

SEE ALSO:

• http://blog.tsunanet.net/2010/11/how-long-does-it-take-to-make-context.html
• https://gist.github.com/jboner/2841832
• http://www.7-cpu.com/cpu/Haswell.html
• http://ithare.com/infographics-operation-costs-in-cpu-clock-cycles/