1 Reply Latest reply on Oct 6, 2016 1:16 PM by Intel Corporation

    Interpreting Intel Memory Latency Checker Output.

    kakarrot

      I am using a dual socket Xeon E5- 2650 machine with 32 GB of memory (Numa Node). I have been wanting to determine

      the memory latency for a CPU located in node 0 and memory allocated in node 1 and vice versa. Below is the output of MLC.

       

      ./mlc  –c0 –i9

      Intel(R) Memory Latency Checker - v3.1a

      Measuring idle latencies (in ns)...

                      Numa node

      Numa node            0       1

             0          69.4   118.4

             1         120.9    69.8

       

       

      Measuring Peak Memory Bandwidths for the system

      Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)

      Using all the threads from each core if Hyper-threading is enabled

      Using traffic with the following read-write ratios

      ALL Reads        :      84419.1

      3:1 Reads-Writes :      78571.6

      2:1 Reads-Writes :      77405.7

      1:1 Reads-Writes :      76944.3

      Stream-triad like:      71137.4

       

       

      Measuring Memory Bandwidths between nodes within system

      Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)

      Using all the threads from each core if Hyper-threading is enabled

      Using Read-only traffic type

                      Numa node

      Numa node            0       1

             0        42956.4 20636.6

             1        20434.2 42859.2

       

       

      Measuring Loaded Latencies for the system

      Using all the threads from each core if Hyper-threading is enabled

      Using Read-only traffic type

      Inject  Latency Bandwidth

      Delay   (ns)    MB/sec

      ==========================

      00000  162.49    84301.1

      00002  161.99    84227.3

      00008  159.84    83922.1

      00015  156.98    83520.4

      00050  111.13    68611.6

      00100   99.14    50781.7

      00200   92.75    33050.8

      00300   89.26    24573.0

      00400   87.29    19625.5

      00500   85.81    16389.1

      00700   82.69    12398.8

      01000   82.53     9169.0

      01300   81.64     7352.0

      01700   80.25     5888.5

      02500   79.06     4325.2

      03500   78.10     3357.7

      05000   77.66     2613.2

      09000   76.99     1832.3

      20000   76.52     1289.1

       

       

      Measuring cache-to-cache transfer latency (in ns)...

      Local Socket L2->L2 HIT  latency        31.4

      Local Socket L2->L2 HITM latency        35.8

      Remote Socket LLC->LLC HITM latency (data address homed in writer socket)

                              Reader Numa Node

      Writer Numa Node     0       1

                  0        -   130.7

                  1    129.4       -

      Remote Socket LLC->LLC HITM latency (data address homed in reader socket)

                              Reader Numa Node

      Writer Numa Node     0       1

                  0        -    80.5

                  1     79.7       -

       

      From this, how can I get the desired latencies?

       

      Thanks.