Specifications

drives in a RAID 0 configuration, the limiting factor should have been the SATA-II bus. However
this was not the case, and sustained transfer rates for the RAID as a whole were limited to about 120
MiB/s. Most of this loss was due to the way Linux caches writes to the drive, but part was due to the
non-optimal factory tuning of the drives’ firmware. When testing the SV35.5 series with the same
options, the performance was approximately the same. However, more recent tests circumvented
kernel cache usage and were able to achieve rates up to 150 MiB/s for several hours. Tests of the
7200.11 series have not yet been performed with this option. However, given the negligible increased
cost of the SV35.5 option, we do not plan to consider the 7200.11 series further.
The test itself consisted of opening several files on the RAID array and writing known data that
was easily verified afterwards. The blocks of data written to the files consisted of either 3900 bytes
or 1008 bytes, the first 8 of which were used as a serial identifier, while the remaining bytes were
filled with an 8-bit counter value, increasing by one for each successive byte and rolling over to 0
after 255. Each time the block was written to the file, the serial identifier was incremented by 1
before writing the block again. This test was performed several times for varying durations, and
the maximum sustained transfer rate was approximately 120 MiB/s when using the kernel cache
for all tests shorter than ten hours. For tests in which the kernel cache was bypassed, a maximum
sustained transfer rate has not been established, though tests indicate this number to be at least
150 MiB/s for all tests shorter than 8 hours.
3.2 File System Performance
The MCS-DR uses a software RAID level 0 array with a custom, raw-mode file system. Other file
systems were considered before selecting a raw file system. Knowing that the journaling operations
of the ext3 file system would require to o much overhead, the first choice of file systems was ext2 for
simplicity. Initial testing had established that the ext2 file system was capable of meeting the 112
MiB/s requirement, but the need to bypass kernel caching made working with ext2 files difficult.
Tests were run with the xfs, fat32
3
, and raw file systems. The xfs test yielded a transfer rate of
75 MiB/s, and the fat32 test resulted in an inconsistent 70 MiB/s, and tests of the raw file system
achieved rates of 150 MiB/s.
3
It should be noted that the fat32 file system would not have supported the full 5 TiB capacity of the RAID array
9