NFS vs NAS vs SAN for Apache?

On Tue, 30 Sep 2003 13:06:49 -0400, Axel Schwenke wrote:


 >
 > But then this central storage will be your single point of failure. Not
 > to mention the possible bottleneck.
 >

Very true. However, my thinking on this is that even though it is,
technically, a single point of failure, it is also a single point of
backup. Instead of backing up 10-20 server I back up one and if something
bad does happen with that one central storage box then I just replace it
and re-load the data.

 > NAS servers are more optimized. But some are just Linux boxes running
 > kernel-nfsd and/or Samba. That need'nt be a problem. @Home my NFS server
 > (Linux) is a PII @300MHz and reaches ~8MB/s in my 100MBit LAN.
 >

Would the performance from a NAS be better than that of a Linux box
serving NFS? If so, would it be substantial? I have been having a hard
time finding _hard_ numbers on performance for NAS devices.

 >
 > I prefer webservers with local storage. At the moment I'm running 12
 > servers. Content is distributed with rsync (over ssh) which is pretty
 > fast and reliable. For comfort I wrote a Perl script that starts rsync
 > with appropriate options in multiple child processes.
 >

Agreed, this is certainly the best option. However, for us we have
several sites running on the web server that are continually updated,
meaning within the span of one minute we may have 100 things change on
the site, mostly images. Using rsync would be a huge nightmare for
someting like this. There is no way we could do it efficently, which is
why I'm thinking the central storage route.

If we ignore the single point of failure aspect for a moment and think of
the performance...

How much of a hit am I going to take serving content from an NFS export?
Currently everything is running on one box with U320 RAID 5. It is fast.
Am I going to see a noticable difference when I connect to the site and
view a page?

I know that is hard to figure out in true numbers, but I'm just trying to
figure out if I'm barking up the right tree or not.

Many thanks.

-- Orca<!-- ~MESSAGE_AFTER~ -->
 >> Stay informed about: NFS vs NAS vs SAN for Apache? 

Orca <bigorca.RemoveThis@hotmail.com> wrote:
 > On Tue, 30 Sep 2003 13:06:49 -0400, Axel Schwenke wrote:
 >
  >> But then this central storage will be your single point of failure.
 >
 > Very true. However, my thinking on this is that even though it is,
 > technically, a single point of failure, it is also a single point of
 > backup.

So what? The only valuable data on a webserver are the access logs.
Content on webservers should be only a copy. Backup your CMS.

  >> NAS servers are more optimized. But some are just Linux boxes running
  >> kernel-nfsd and/or Samba. That need'nt be a problem. @Home my NFS server
  >> (Linux) is a PII @300MHz and reaches ~8MB/s in my 100MBit LAN.
 >
 > Would the performance from a NAS be better than that of a Linux box
 > serving NFS?

It depends.

1. There is not "the" NAS box
2. There is not "the" Linux box
3. Administrator skills vary

 > If so, would it be substantial?

Possibly not.

 > If we ignore the single point of failure aspect for a moment and think of
 > the performance...
 >
 > How much of a hit am I going to take serving content from an NFS export?
 > Currently everything is running on one box with U320 RAID 5. It is fast.
 > Am I going to see a noticable difference when I connect to the site and
 > view a page?

What is the total bandwith of your webservers? An U320 RAID5 can
deliver $NUMBER_OF_DISCS-1 times the bandwidth of a single disc or
~300MB/s (whatever is less). A NAS box' bandwidth is limited by a
few more things:

1. Storage inside the box
2. Network bandwith between box and client(s)
3. Processor power @box and @client(s)
4. Protocol overhead, protocol implementation costs (low for FTP,
medium for NFS, high for SMB)

IMO it will be hard to outperform your "local RAID-5" configuration
with a NAS box. Just the network (can you afford something faster than
Gigabit-Ethernet?) will limit your bandwidth to 100MB/s.

Sounds a bit theoretical now. Do you really deliver that much content?


XL
--
Das ist halt der Unterschied: Unix ist ein Betriebssystem mit Tradition,
die anderen sind einfach von sich aus unlogisch. -- Anselm Lingnau<!-- ~MESSAGE_AFTER~ -->
 >> Stay informed about: NFS vs NAS vs SAN for Apache? 

schwenke RemoveThis @jobpilot.de (Axel Schwenke) wrote in message
  > > Very true. However, my thinking on this is that even though it is,
  > > technically, a single point of failure, it is also a single point of
  > > backup.
 >
 > So what? The only valuable data on a webserver are the access logs.
 > Content on webservers should be only a copy. Backup your CMS.
 >

Huh? That is a very odd statement and it confused me. Web servers
contain the data (files, etc) that they are being served to the world.
The bulk of the actual 'data', meaning text, etc lives in the
database, which is fine.

We are talking about storage of actual files that are being served.
That is what I am interested in at the moment. I'm assuming you run a
setup where files are stored elsewhere and copied to your web servers?
As I stated before, this will not work for us as the amount of data
that changes on a minute by minute basis is quite large and it would
be near impossible to keep up with syncing boxes.


  > > Would the performance from a NAS be better than that of a Linux box
  > > serving NFS?
 >
 > It depends.
 >
 > 1. There is not "the" NAS box
 > 2. There is not "the" Linux box
 > 3. Administrator skills vary
 >

True, very valid point.


 > What is the total bandwith of your webservers? An U320 RAID5 can
 > deliver $NUMBER_OF_DISCS-1 times the bandwidth of a single disc or
 > ~300MB/s (whatever is less). A NAS box' bandwidth is limited by a
 > few more things:
 >
 > 1. Storage inside the box
 > 2. Network bandwith between box and client(s)
 > 3. Processor power @box and @client(s)
 > 4. Protocol overhead, protocol implementation costs (low for FTP,
 > medium for NFS, high for SMB)
 >
 > IMO it will be hard to outperform your "local RAID-5" configuration
 > with a NAS box. Just the network (can you afford something faster than
 > Gigabit-Ethernet?) will limit your bandwidth to 100MB/s.
 >
 > Sounds a bit theoretical now.

It is totally theoretical at the moment. I would hate to rush in to
building something out without thinking about it!

I wouldn't expect to outperform our current RAID 5. I expect it to be
slower. The whole idea here is to end of with multiple front-end
servers that are load balanced and using a single storage area. This
makes our front-end boxes 'disposable' if you will. If one dies it
won't matter. We just replace it when we can.

I was just trying to find out if anyone else has ever done anything
like this. I believe Slashdot does something similar, but I'm not sure
exactly how their setup is.

 > Do you really deliver that much content?

Yes.


Thanks for your help. I appreciate your thinking on this and would
love to hear anything any has to add.

Thanks again!

-- Orca<!-- ~MESSAGE_AFTER~ -->
 >> Stay informed about: NFS vs NAS vs SAN for Apache?