[Gluster-users] Performance

[Todd Daugherty]

the right benchmark is my actually workflow I have benchmark of all of that.

ib_write_bw -a 192.168.0.112
------------------------------------------------------------------
                    RDMA_Write BW Test
Number of qp's running 1
Connection type : RC
Each Qp will post up to 100 messages each time
Inline data is used up to 0 bytes message
  local address:  LID 0x0d, QPN 0x24004f, PSN 0x1d2b7a RKey 0x8045c00
VAddr 0x007fc23a9c8000
  remote address: LID 0x09, QPN 0x24004f, PSN 0xda5165, RKey 0x8045c00
VAddr 0x007fbf89eea000
Mtu : 2048
------------------------------------------------------------------
 #bytes #iterations    BW peak[MB/sec]    BW average[MB/sec]
      2        5000               8.01                  7.98
      4        5000              16.07                 16.04
      8        5000              32.04                 31.97
     16        5000              63.99                 63.91
     32        5000             128.58                128.37
     64        5000             256.35                255.77
    128        5000             437.59                437.04
    256        5000             812.91                812.12
    512        5000            1524.92                835.86
   1024        5000            2776.72               1645.07
   2048        5000            3085.98               3084.23
   4096        5000            3163.76               3163.50
   8192        5000            3209.57               3209.07
  16384        5000            3234.23               3234.02
  32768        5000            3243.04               3242.91
  65536        5000            3248.66               3248.63
 131072        5000            3251.42               3251.36
 262144        5000            3252.86               3252.84
 524288        5000            3253.60               3253.59
1048576        5000            3253.95               3253.94
2097152        5000            3254.14               3254.14
4194304        5000            3253.59               3253.59
8388608        5000            3253.43               3253.43
------------------------------------------------------------------

Fabric is fine and fast. I am posting because I have complete
confidence that everything else but GlusterFS is doing what it is
supposed to be doing. Is there anyone out there that is moving large
and amounts of data to a single client at a high bandwidth?

Single stream performance on QDR anyone?

Is anyone getting a single stream read performance above 2.5 gigabytes
per second? If so how?

Todd