304
The Intel Memory Latency Checker (Intel MLC) is a free tool for Linux and Windows
available from https://software.intel.com/en-us/articles/intelr-memorylatency-checker. Intel MLC can be used to measure bandwidth and latency of DRAM
and persistent memory using a variety of tests:
• Measure idle memory latencies between each CPU socket
• Measure peak memory bandwidth requests with varying ratios of
reads and writes
• Measure latencies at different bandwidth points
• Measure latencies for requests addressed to a specific memory
controller from a specific core
• Measure cache latencies
• Measure b/w from a subset of the cores/sockets
• Measure b/w for different read/write ratios
• Measure latencies for random and sequential address patterns
• Measure latencies for different stride sizes
• Measure cache-to-cache data transfer latencies
VTune Profiler has a built-in kernel to measure peak bandwidth on a system. Once
you know the peak bandwidth of the platform, you can then measure the persistent
memory bandwidth of your workload. This will reveal whether persistent memory
bandwidth is a bottleneck. Figure 15-6 shows an example of persistent memory read and
write bandwidth of an application.
Figure 15-6. Results from VTune Profiler persistent memory bandwidth
measurements
Chapter 15 profiling and performanCe
The Intel Memory Latency Checker (Intel MLC) is a free tool for Linux and Windows
available from https://software.intel.com/en-us/articles/intelr-memorylatency-checker. Intel MLC can be used to measure bandwidth and latency of DRAM
and persistent memory using a variety of tests:
• Measure idle memory latencies between each CPU socket
• Measure peak memory bandwidth requests with varying ratios of
reads and writes
• Measure latencies at different bandwidth points
• Measure latencies for requests addressed to a specific memory
controller from a specific core
• Measure cache latencies
• Measure b/w from a subset of the cores/sockets
• Measure b/w for different read/write ratios
• Measure latencies for random and sequential address patterns
• Measure latencies for different stride sizes
• Measure cache-to-cache data transfer latencies
VTune Profiler has a built-in kernel to measure peak bandwidth on a system. Once
you know the peak bandwidth of the platform, you can then measure the persistent
memory bandwidth of your workload. This will reveal whether persistent memory
bandwidth is a bottleneck. Figure 15-6 shows an example of persistent memory read and
write bandwidth of an application.
Figure 15-6. Results from VTune Profiler persistent memory bandwidth
measurements
Chapter 15 profiling and performanCe