Some potential usages of mogilefs

Brad Fitzpatrick brad at
Mon May 21 15:08:41 UTC 2007


I often wonder if you ever try anything, or just ask questions on the
mailing list, but I'll continue to humor you with answers.

Lighttpd involves:

   -- getting request for /foo/img.jpg and returning it

MogileFS involves

   -- getting request for /foo/img.jpg,
   -- asking mogilefsd where /foo/img.jpg is, which;
      -- finds out that /foo/img.jpg is currently fid #123
      -- returns that fid #123 is on device 5 and device 7,
         returning both URLs
   -- proxying the request for /foo/img.jpg to

Which looks like more overhead to you?

Where MogileFS wins is that you spread out your storage, and things are
constantly replicated/safe if a single machine or host dies.

That said, there's a lot you can do to make the above steps fast/cached.
If you use Perlbal as your front-end load balancer, the part of your app
that looks up /foo/img.jpg -> FIDs(5,7) can, when returning the list of
FID URLs to Perlbal, instruct Perlbal to remember that mapping for 'n'
seconds, so when the next /foo/img.jpg request comes in, probably in under
a second on popular sites, all those steps are skipped and the request
goes right to the mogile storage node (which can be RUNNING lighttpd

So MogileFS can be fast, but it's always going to have at least that one
more step, sometimes two, so don't expect lighttpd speeds.  We care about
scalability and uptime around here, not raw speed on a single node with
ideal conditions.

Use MogileFS for the right reasons.

- Brad

On Mon, 21 May 2007, howard chen wrote:

> Hello,
> 1. Do you think mogilefs for serving large amount of static files,
> rarely changed, e.g. images, javascript, css etc.
> Somethings similar to Yahoo's yimg (but not yahoofs).
> Currently we are using load balancer + rsync, which suite our need if
> the scale is small, but in the future, if you have large amount for
> files, rsync will become the bottleneck
> 2. Are there any performance metrics available? such as max. of
> storage volume in real world setup, number of requests/response per
> seconds on a modern server...I know it is difficult to compare as
> people always has different setup, but I would like to know more abt
> the general overhead, e.g. serving 1MB JPEG file from lighty (1.5,
> async) vs mogilefs
> Thanks for any inputs...
> Howard

More information about the mogilefs mailing list