Watching performance on a DHCP Server
Simon Hobson
dhcp1 at thehobsons.co.uk
Tue Feb 12 22:17:19 UTC 2008
Olaf van der Spek wrote:
> > fsyncs are the most I/O intensive operation you can perform on magnetic
>> media. If I remember correctly it takes at least 2 full rotations of the
>> disk for an fsync to occur, although it may take longer to commit more data.
>> This is were solid state rules over magnetic storage.
>
>Why are two required?
>Even then, you could commit lots of transactions in a single go.
I'm wondering if that's a hangover from RAID5 thinking - where
(unless the data is already in cache), every write is preceded by a
read. In fact, to write a single block, the minimum is to read the
pre-update version of that block and the pre-update parity block that
goes with it, re-compute the parity, and write the new data and
parity. In that situation, you must read and then write the same
block, so a minimum of just over one revolution, with an average of a
little over 1 1/2 revs - all assuming zero overhead in computing the
new parity and synced spindles.
More information about the dhcp-users
mailing list