siberian wrote:
> 
> Hi All-
>         I am building a pretty in depth architecture for our new service
> using ModPerl. I've done a lot of large scale/high traffic apps in modperl
> before but never in conjunction with a network attached file server. I am
> thinking that it would really make my life easy to have one central
> repository of code, databases and sundry files that all the servers share
> ( making it easy to swap out servers, add servers etc since its a central
> file repository that everyone just hooks into ).

Sounds like you are building in a single point of failure.  If your
NetApp goes down, everything goes down.  Do you really have so much data
that you can't easily synchronize the hard disks of several machines?

If you do go with a NetApp, you need to think about it a lot.  NFS has
issues.  For example, let's say that your NFS client software has a
timeout period of 120 seconds.  Let's also assume that each of your
machines is serving 10 requests per second with 10 apache processes and
is using half of its RAM.  Suddenly, the NetApp fails.  Horror of
horrors!  During your 120 seconds, 1200 requests queue up, which would
require 1200 Apache processes to be blocked doing nothing.  If your
Apache is configured correctly, these people will wait forever for error
messages.  If Apache is not configured correctly, your machine will run
out of swap and the kernel will panic.  Neither of these is a good
thing.  Compare and contrast with a local disk failing on one server. 
The failure only affects a fraction of requests, and I/O errors on the
local subsystem are usually quite quickly reported.

The other thing to consider is that, if the NetApp goes down, it will be
down for a long, long time.  Their fsck isn't any faster than, for
example, Linux's fsck.  If the NetApp is very large, this could take
hours.  If you think the NetApp can't fail, I'll tell you that I've seen
several failures in the real world.  I've even seen them get so fucked
up that they couldn't complete their fsck, after of course thrashing for
16 hours.  Some of these machines had extremely demanding I/O loads, but
others were simple file servers supporting a development organization.

That's my take.

-jwb

> My question is : Has anyone experienced any 'gotchas' in putting perl code
> that modperl handlers use on a Network Attached file server like a network
> appliance box ( www.netapp.com )? I am assuming that there are no real
> issues but before i go blow a ton of cash on this thing I wanted to be
> sure that no one had found a problem.
> 
> Thanks for any pointers.
> John Armstrong

Reply via email to