Posted Feb 7, 2008 4:15 UTC (Thu) by jwb (guest, #15467)
Parent article: CRFS and POHMELFS
It's a bit odd to publish performance data for software which has not even reached the level
of half-baked yet, isn't it? One can easily imagine how trivial implementations are fast.
See /dev/zero. To benchmark a coherent remote file system when the coherency is purely
theoretical strikes me as premature. It's the same with any new filesystem when the author
gets on l-k and states that it's ten times faster than ext3, but, by the way, it doesn't
implement rename, remove, or hard links (yet). Eventually the performance worsens as data
structures and code are added to deliver required features.
If you are actually interested in the performance aspects of coherent network file systems,
there are a number of implementations which have existed for many years. There is also the
newer, more vaporous pNFS effort.