[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: swaps and partition mounts
Yeah, besides suggesting memory or a solid state SCSI/HIPPI device,
I/O performance is really dependent on the situation. You can get
better performance using outer cylinders, but that gets washed out in
most cases with the disk buffer, OS buffering, access patterns,
etc. Pretty much any general scheme you come up with I bet someone can
respond with an I/O situation that will trash your performance.
You can burn endless hours tuning these things for very little gain. I
always start with the big requirements and then drill down: what is
running on the machine (besides Apache in this case) -> what are the
access patterns for each application (small webpages several times a
second?, scripts/php running with each access?) -> how do I make
those access patterns mesh with what I've got. You also need to
remember that for a webserver your key bandwidth limiter is
probably going to be your network connection. Why worry about
that last 10% of performance if your machine can already serve
out pages 50 times faster than anyone can request them?
For the website I work with, ease of use is more important than
absolute access speed. I set the developers up with CVS and then gave
them two mount points -- and active one and inactive one. Whenever
they're ready for a change, they do a CVS export to the inactive one
and then swap the link to make it active. This implementation factor
was much more important than performance concerns. For that I just
stuck the whole thing on a RAID5 and said good enough.
Anyway, that's my take....
David S. writes:
> > The current question being explored is for servers that
> > run dedicated services such as web, mail or whatever
> > and whether using partitions for its various functions.
> > Lets assume its apache on OpenBSD3.0. Does anyone
> > have any emotions towards running apache on its own
> > partition and the webcontent on another partition
> > or something similar? emotions on i/o resonse speed
> > on a partition mounted on '/' vs multiple mount
> > locations?
> Putting your web content on a separate partition on the root disk
> won't do much for I/O performance (OK, I have read reports here of
> the outer cylinders of some disks being faster or slower than the
> inner), but it would prevent an errant process writing to the web
> directory from filling up that disk. If you want better I/O, put
> the web content on a separate spindle; even better, on a separate
> spindle on a separate SCSI controller; better yet, striped across
> two spindles; more, striped across two spindles on two separate
> controllers ...
> Want even better I/O performance for static web content? Here's
> a crazy idea (which I haven't tried, and which somebody here may
> shoot down as unfeasible/impossible/stupid): stuff your system
> with as much RAM as you can, and serve up your HTML out of the
> memory file system.
> David S.
Jonathan D. Dunfee