On 08/08/2009, Matthew Dillon <[email protected]> wrote: > It will depend on what kind of test you are running. Anything > that calls fsync() is going to be fairly horrible for HAMMER... > flushing the filesystem is very expensive and going to remain > that way for a while. > > Normal sequential write I/O without fsync() should be fairly close > to platter rates, at least that is what I get, e.g. with something > like dd. Sysbench, and yes, it calls fsync. I get about half of the UFS performance data. dd is OK.
The performance is also lower when reading, here are some lines from the random read test: dfly hammer: #bklsize[kB] MBps iops min avg max 95% - latency times in ms 1 .14 150.77 0.0000 0.0066 2.4605 0.0101 2 .34 177.72 0.0000 0.0056 1.5467 0.0083 4 .64 165.02 0.0000 0.0060 2.7946 0.0087 8 1.39 179.03 0.0000 0.0056 0.0179 0.0084 16 2.59 166.17 0.0000 0.0060 4.1047 0.0086 32 4.81 153.94 0.0000 0.0065 0.8895 0.0103 64 10.59 169.49 0.0000 0.0059 2.5563 0.0085 128 15.54 124.35 0.0001 0.0080 3.3134 0.0117 256 29.90 119.61 0.0001 0.0083 2.2407 0.0116 fbsd8 UFS: #bklsize[kB] MBps iops min avg max 95% - latency times in ms 1 .20 212.32 0.01 4.71 833.94 7.62 2 .40 206.95 0.01 4.83 321.63 7.84 4 .84 216.99 0.01 4.60 1751.44 7.47 8 1.65 211.71 0.01 4.72 321.55 7.66 16 3.34 214.37 0.01 4.66 245.09 7.54 32 6.34 203.13 0.01 4.92 570.37 7.88 64 11.66 186.58 0.02 5.36 516.09 8.40 128 20.30 162.41 0.04 6.15 517.97 9.41 256 34.13 136.54 0.07 7.32 667.05 10.74 This is with a P400, 512 MiB BBWC and a 15k RPM SAS disk. -- http://suckit.blog.hu/
