Date: Tue, 21 Mar 2000 11:37:24 +0100 From: Thierry.Herbelot@alcatel.fr To: b.j.smith@ieee.org, thebs@theseus.com Cc: freebsd-questions@FreeBSD.ORG, freebsd-hardware@FreeBSD.ORG Subject: Re: Looking to use FreeBSD as an NFS/App server for 10 Solaris clients ... Message-ID: <C12568A9.003A5C6B.00@frmta003.netfr.alcatel.fr>
next in thread | raw e-mail | index | archive | help
Hello,
(see answers interleaved with your questions)
TfH
"Bryan J. Smith" <b.j.smith@ieee.org> on 21/03/2000 11:17:02
Please respond to b.j.smith@ieee.org; Please respond to thebs@theseus.com
To: freebsd-questions@FreeBSD.ORG,
freebsd-hardware@FreeBSD.ORG
cc: (bcc: Thierry HERBELOT/FR/ALCATEL)
Subject: Looking to use FreeBSD as an NFS/App server for 10
Solaris clients ...
Looking to use FreeBSD as an NFS/App server for 10 Solaris clients
...
[SNIP]
Four questions:
===============
I. Looking to install from the new 4.0-RELEASE ISO
Comments versus 3.4-STABLE?
II. Cost aimpoint is only $1,000 due to budget
(2)Cel466+256MB+(2)30GB-7200+Tulip will meet that
Comments on SMP, RAM size, IDE, Tulip?
III. Software RAID-0 for speed? Installer, tools ...
IV. Sole purpose, NFS v3 server to Sol2.6 clients
~28GB App/ROdata and using 4GB CacheFS on clients
Comments on compat with Solaris?
V. NIS/YP server is Linux. No shadow. Comments?
I. Install from 4.0-RELEASE ISO
================================
I am looking to install my first production FreeBSD system from
4.0-RELEASE stable. The main reason is that I have heard the NFS
v3
and locking is very mature. Secondly is the fact that SMP is
------------
NFS locking is not yet part of any version of FreeBSD
SMP on FreeBSD will give you nothing if your main load is I/O
(only one processor can do I/O tasks at one given time)
------------
supposively quite stable. Lastly, the feature list is nice.
Based on the hardware, use and other criteria below, am I better
off
with 3.4-STABLE, or will 4.0-RELEASE be either better or not any
less stable?
II. System is $1,000 due to budget
===================================
I've priced it out and have managed to get a total of $996.xx
(including shipping, under the $1,000 mark before I have to kiss a
lot of @$$) from my favorite reseller (mwave.com) for a Abit BP6
dual-Celey, (2) Celeron 466MHz processors, a generic 256MB PC100
DIMM (only runs at 66MHz in this config), (2) Maxtor DiamondMax
Plus
30GB 7200rpm ATAPI drives (40GB series, 3 platter), a Linksys
10/100
card (Tulip) and a 4MB ATI AGP Rage IIc card. I will buy the
enclosure separately (looking a few $100-130 good airflow
enclosures
from SuperMicro, or my favorite enclosure reseller, directron.com).
The mainboard/CPU combo. I have been using this combo in both home
and production Linux boxes. I always get the retail Celeys to make
sure. MWave is always good to give me the same
stepping/sub-stepping too. Any comments on this under FreeBSD 4.0
(or 3.4)? How's the SMP support? Should I expect to have to
recompile the kernel to get it (not hard, just new to doing it in
FreeBSD)?
The memory is 256MB. I feel that should be sufficient if it's
running headless (no GUI, etc...). Basically there for the OS, the
NFSd's and caching. Any comments?
The hard drives are Maxtor DiamondMax Plus 40-series, only 3
platters instead of 4 (30GB). These babies are only $209 at
Multiwave! Personally, I _love_ Maxtor's ATAPI drives, and so does
StorageReview.com. They rank very high on NT benchmarks (not so
much Win9x), and the Adaptec ThreadMarks (Maxtor's 5400rpms often
beat out other vendor's 7200rpms). Personally, I think Maxtor has
a
nice command chipset that gives it some SCSI-like features for
------------
Soren (the FreeBSD ATA-guru) generally recommends IBM drives, as they
more closely follow the IDE standards - YMMV
------------
low-multiuser apps (like power workstations or little workgroup
servers, etc...). I have gotten _awesome_ performance out of them
in Linux on my web, ftp and other Internet content servers. Now I
would still use SCSI in a full-up, main file server, this system
only serves one purpose (see IV below). I think it'll work for
what
I need without taxing the CPU much. I am going to _disable_ the
UltraDMA/66 support (down to UltraDMA/33, done it before on my
Linux
boxes), and use the native PIIX4e southbridge on the BP6. Does
FreeBSD do full UltraDMA/33 (I assume so)? Or does it even support
the HPT366 controller and UltraDMA/66 now?
Lastly, any comments on the Linksys card? It's a recent Tulip
variant, probably the Lite-On and I have used them with great
success. I was a Intel EEPro100+ fan until the transmitter
starting crapping out my main Linux fileserver (stupid i82558,
although my older 100B i82557s seem to be fine).
III. Software RAID-0 for speed? Installer, tools ...
======================================================
I would like to go software RAID-0 for speed, alongside using
softupdates. Since my files will be largely static (see IV.
below),
I think this will be fine. Does the 4.0-RELEASE installer have
RAID-0 setup in the setup? Or will I need to do it in a
post-install? The later will be fine, and I'll just stripe the
data
partitions (as one big ~50GB one).
------------
vinum(4) will do exactly what you want (I'v got two 10G IDE drives at
home striped into one partition)
------------
Secondly, how are the tools? I am thoroughly upset with Linux's
tools, at least for RAID-1. If you disconnect one drive to "test"
your LILO config/script to make sure the 2nd drive can boot, when
you reconnect it loses the spare! Ouch! There seems to be no way
to manually restore RAID-1 (RAID-5, yes, but not for RAID-1 and the
automatic scripts seem to miss it). But I'm no expert.
Any other comments?
IV. Sole purpose, NFS v3 server to Sol2.6 clients
==================================================
Okay, this is the _meat_n_potatoes_. Before I got to my employer,
there was (and still is) a bit of cross-mounting going on between
the Sun workstations. Generally, a no-no in my book. And as users
slam the systems and hit 100% utilitzation and 100% VM usage
(2GB+!), the NFS/network performance stalls. These cross-mounts
are
for app sharing.
I am trying to centralize these apps because every box needs to run
every app, and the total of all binaries we use are ~28GB -- so
local is out-of-the-question (let alone a headache on 10 systems!).
I would normally use one of the Sun boxes, but I am running up
against that fact that A) I need all the SPARC CPUs I can get (At
least until more EDA tool vendors port to Linux = cluster time!
And
I've done it with Aerospace apps before too!), and B) the 30GB+ in
external SCSI drives would probably cost me >$1,000.
So I arrived at the FreeBSD headless box solution as described in
above. The data will be largely static, apps only. I am also
going
to setup a 4GB CacheFS file on each Solaris box (except two, which
will only have 1GB due to disk contraints) to reduced network
traffice.
I need to know how stable FreeBSD's NFS v3 is for Solaris 2.6
clients. In 4.0-RELEASE? How about 3.4-STABLE?
------------
the best NFS implementation should be on 4.0 (however, this is not
*yet* recommended for production)
------------
[SNIP]
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?C12568A9.003A5C6B.00>