[nflug] NFS isn't (all) bad.

Thu Apr 27 17:47:43 EDT 2006

There's been a statement put up by an individual "That NFS is bad,  I
won't use it, it's so out-dated and so full or problems" and so on. 
The thread started because someone was having some NFS issues and it
related to a mail server and mail spools or homedirs residing on an NFS
mount and some difficulty arrived.  I presented a tiny bit of evidence
about NFS performance as did some others about it's merits.  The
response I got seemed completely off topic, so I wanted to show some
examples where NFS is used in the corporate world, and if it's not
considered optimal,  I'd like to see a "better way" to do it.  AS I'm
always looking to learn new stuff.

Example 1. NFS for remote bootable clients.
Lets say you have a school. Schools are notoriously low budget , thanks
to many things these days, so some places are turning to using thin
clients in their computer labs, and Linux is ideally suited for the
task (free, easy to configure, very flexible, wide variety of
choices/applications).  You have a stack of old PC's (let's day
PII-400's), you have next to no budget for any replacement parts like
disks and half your machiens are in the 5-8 year old range and have
dead drives, what to do?  Remote boot them, either via a etherboot
floppy, or via a PXE Nic to a central cerver, using software like
PXE's, LTSP (Linux Term Server Project), or Knoppix Terminal server. 
It's simple,  but all of them rely on NFS as shared filesystem, why? 
because it's 
1. Clean (simple to setup) 
2. Easy
3. Built into the OS distribution, 
4. (optional, not always used) The kernel can directly use a NFS
filesystem for it's root FS,  it cannot do this with ANY other network
filesystem without patching of the kernel (which is hard for someone
who hasn't done it before)

Without using NFS these projects would be stunted and held back.  NFS
(as far as I can see it) is the optimal tool for the job in the above
case.

2. Corporate home dirs (unix shop). 
One of the places I worked (Valeo, Rochester, 4+ years) was a mixed bag
shop,  windows AD, Unix, Linux etc.  They had a large number of unix 
users, about 130 spread across two locations across a WAN link (SGI's
mostly, though some HP and linux)  All of the Unix hosts made use of
shared storage locations available via NFS from a Network Appliance
Filer (big fileserver appliance that was used fot NFS and samba).  The
users used it for homedirs, application dirs,etc and even mounted
locations from over 250miles away (detroit area) via NFS over the WAN
link, and it worked flawlessly. 

3. Small business backup server
 Another place I've doen a little helping out from time to time, is
asuccessfull computer business,  They have a server with a bunch of
disk tucked away in some hidden part of their office, and a full
netwrok throughout.  All servers backup to it via a cron job dump based
system (using "dump" to dump to files stored on the NFS server).  It
work great, is fast, allows easy recovery anyplace on the LAN and keeps
things centralized.  NFS in this case makes the filesystem essentially
transparent.  They have a large pool of diskspace that can be mounted
anywhere anytime on the LAN.

4. Remote server installation
I had to do an install of 9 mid-High end Linux servers (4-16CPU, up to
64GB ram boxes) as part of my current job earlier this month.  They're
a Redhat Enterprise shop, and I was limited by them to using an older
version (EL3) Their constraints were simple. Get it done fast, and get
it done right. IT helped that they required minimal customization. The
fastest way I know to get a RH Box built is by using Kickstart, and a
remote boot.  Remote boot wasn't possible for their datacenter, as
DHCP/Bootp was not allowed, so I used the floppy method (faster than
burning up custom CD's) Also note that these servers were located north
of Boston Mass, and I was still in buffalo and everything needed to be
done via remote. (had a guy onsite to insert floppies for initial
bootup). A NFS server was setup (super EASY), as every linux box that
had it installed via default as it was IN USE there. The ISO images for
RHEL3 was put in place as well as kickstart config files tailored to
each machine. I had a tech boot the machines off of floppy, while I
watched them through a remote IP KVM (two disks were needed to do the
NFS boot (the drive disk was necessary for gigabit ethernet support). 
>From the point where the second floppy was read and the install started
to the time it finished and the machine was ready to reboot took 13
minutes installing via NFS. It installed about 3GB worth of data, so it
wasn't an "everything" install.  Yes I could have done an FTP or HTTP
install,  but those require significant more setup time, as there were
not FTP servers or http servers available for use for that task,  but
nearly ever unix/linux box in the datacenter (and there's about 185
systems in there) could have been used for NFS.

Hope you all enjoyed my long winded uses of NFS.  If anyone knows a
better way that's more efficient, uses less CPU and is easier to do, 
please respond.

So NFS is used in the real world, even though it's been around a long
time (which is a GOOD thing),  it's alive and well, and has more uses
than you might think.

Apologies for all the spelling and gramatical errors.

-- David Andruczyk
Unix/Linux Systems Administrator
IBM Global Services

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
_______________________________________________
nflug mailing list
nflug at nflug.org
http://www.nflug.org/mailman/listinfo/nflug