We tried something simillar in our shop.
-------------------------
Item #1:
We tried to find a "cheaper" alternative to
EMC Celerra. We tried nice 4 processor box under
RedHat, with kernel 2.4, of whatever version was
a last one last spring. We did not try 2.5...
The setup was;
40 clients. Connected via 100 full duplex. They
did a really large build (many thousands of files)
on NFS-mounted partition, shared from the server
via copper gigabit. The network was CISCO
Catalist, with 4 gigabit backplain. Storage was
via SAN, 8 or 12 drives, stripped accross.
One build was actually tolerable, but the second
one did slow things down. NFS-2 or 3 did not make
a difference. We had all problems which you are
having.
We tried also SUN, Cherrystone, with 2GB of
memory. Solaris 8, pathched, with pumped up number
of nfsds. Same storage, and fiber gigabit to a
switch.
Same build ran OK, and secong build lowed things
down by about 20%. No errors at all.
The verdict was that Linux does not scale really
up, as far as NFS goes.
---------------------------
Item #2:
Are all your drives have the same problem with
vanishing bad sectors? I saw same on first 120GB
drives, when they only came out. Turn out to be
just bad plates, magnetic density was too high for
head mechinism, which failed on positioning. As a
result, sequential reads were OK, but if yuou jump
there and back, you get an error. We just replaced
drives, until we found those which worked.
On Fri, 7 Mar 2003, Doug Mildram wrote:
> Hi ... ! Is anyone interested in sharing progress,info
> (or conceivably contracting for a few days w/your expertise)
> on getting redhat7 to scream as an NFS server? (project 1).
>
> Sorry this is a bit long. There's actually 2 puzzling things i'm chasing:
>
> -------details on project 1, redhat 2.5 kernel, disk I/O, NFS related:
>
> I've got a dual-Xeon w/ 2 "3ware" hw-raid controllers, w/IDE disks.
> They appear as a big pair of 1tb SCSI disks. The box also has a
> well-behaved gb-enet NIC, and the need for good NFS server performance.
> We spent ~$13k, we can't buy NetApp for EVERYTHING, and this particular
> need for "engineer scratch data" isnt "vital" or backed up.
>
> the 2.4 kernel bogs down (the load goes high, higher than the # nfsd's);
> clients suffer w/"nfs server not responding", though it does keep working.
>
> I think the kernel is the answer. I read that 2.4 I/O can't use HIGHMEM
> ( LOWMEM only goes up to 900mb ), I also read that 2.5 has fixed that,
> and I'd like this beast to perform better (and use its full 4gb RAM).
>
>
> So I'm running 2.5 (as of yesterday).
> But it's a fresh/dangerous work in progress...seems faster maybe, but
> it still gets too slow. Also, I havent got both CPU's recognized yet,
> and I havent added my 3rd and 4th GB of RAM yet. Am I the only one (of us)
> nuts enough to try the 2.5 kernel for something like this?
>
> ----------non-contract-worthy project/question 2:
>
> On simple IDE, single disk Redhat7.x/2.4 systems,
> is anyone getting "I/O errors" on the hard disk ?
>
> SYMPTOM: i'm seeing on MULTIPLE (some of my) Tyan/Athlon systems
> which I've had for a few months, are generally good, but some suffer....
>
> -- /dev/hd? mounted on /tmp becomes read-only,
> -- console echos "bad sector" errors, which later prove to be not bad;
> -- entire system becomes unable to log to /var, or behave reasonably;
> -- on reboot, all is well again....for a few days or weeks.
>
> Unsure if the disk SIZE (160gb or 200gb) has anything to do with this;
> the problems pop up on the active, "small" (under 10gb) filesystems.
>
> The disks (WDC WD2000JB/200gb or Maxtor 4G160J8/160gb)
> then plead innocence (pass standalone-floppy-diags from vendor),
> and a reboot "fixes" the problem. My vendor is sucking wind on this,
> meanwhile I am blindly trying OTHER disk models.
> Barking up the wrong hardware tree ?
> I've already tried the best-recommended 2.4 kernel 2.4.18-18.7.x .
>
> ======== have a great day, thanks for listening! --------doug
>
>
> ---
> Send mail for the `bblisa' mailing list to [EMAIL PROTECTED]'.
> Mail administrative requests to [EMAIL PROTECTED]'.
>
--------------
Dimitri (Dima) Shcherban
Phone: 800-445-2588+3+36955
or: 508-898-6955
Westborough ext: [823]-36955
Cellular: 508-889-4918
Pager: 877-563-1780
email: [EMAIL PROTECTED]
Pager email: [EMAIL PROTECTED]
---
Send mail for the `bblisa' mailing list to [EMAIL PROTECTED]'.
Mail administrative requests to [EMAIL PROTECTED]'.
