Post #51,069
9/8/02 10:16:11 PM
|
ooh ooh ooh I know!!!!!
Are you using a mix of oS? then locking is fscked unless params are set on both sides. Also even between two suns, specifying the packet params on both ends works best.
thanx,
bill
will work for cash and other incentives [link|http://home.tampabay.rr.com/boxley/Resume.html|skill set]
[link|http://home.tampabay.rr.com/boxley/index.html|boxley's home page]
qui mori didicit servire dedidicit
|
Post #51,120
9/9/02 1:27:20 PM
|
No locking
All processes write to their own files.
Client is Linux, server is Solaris 8.
Any particular params I should be aware of?
|
Post #51,147
9/9/02 6:48:07 PM
|
yer locking via nfs
when you open a file for a write on a nfs mounted drive, the machine that is local to the write invokes network locking. The sun expects it and the linux prior to 2.4 doesnt implement it so some thrashing is involved. I remember I had the best luck with the rsize and wsize being explicitely set to 8192 on both sides.
thanx.
bill
will work for cash and other incentives [link|http://home.tampabay.rr.com/boxley/Resume.html|skill set]
[link|http://home.tampabay.rr.com/boxley/index.html|boxley's home page]
qui mori didicit servire dedidicit
|
Post #51,149
9/9/02 7:22:48 PM
|
I'm confused
I thought applications issue lock ioctls.
I was unaware of automatic locking except
for Samba.
Could you give me a pointer (google, etc)
for me to read up on the subject.
|
Post #51,162
9/9/02 9:15:23 PM
|
yer right but lockd isnt running on linux
had a full page of links that just blew up on me. So here is one that might help
[link|http://groups.google.com/groups?q=linux+solaris+nfs+performance+issues&start=20&hl=en&lr=&ie=UTF-8&selm=acdjb6%24o7i%241%40news.duke.edu&rnum=29|link] and they suggest nfs.sourceforge.net also
thanx,
bill
will work for cash and other incentives [link|http://home.tampabay.rr.com/boxley/Resume.html|skill set]
[link|http://home.tampabay.rr.com/boxley/index.html|boxley's home page]
qui mori didicit servire dedidicit
|
Post #51,172
9/9/02 11:22:10 PM
|
Yup, my issue
Thanks Box.
I forwarded the thread to my sysadmin.
|
Post #51,219
9/10/02 12:00:39 PM
|
Piece o'crap Dell Dual Xeon
Motherboard based tigon GB chipset.
Driver does not seem to handle dual interrupts
from the dual CPUS.
Put an Intel GB card in, performance skyrocketed.
|
Post #51,253
9/10/02 3:57:33 PM
|
Re: Piece o'crap Dell Dual Xeon
What'd I tell you... Barry... ;)
I build them or spec them... Exactly why I usually avoid this kind of thing...
I usually build everything I can, or get someone to build it for me...
You MAY want to look at Microway or Penguin Computing for these kinds of things... they really do indeed rock...
Just a thought... Watch out for that E1000 driver in RH-7.3 it has some re-transmission problems on re-direct... big ones... if you replace the ".c" and the ".h" in the directory in the redhat source tree /usr/src/linux-2.4.18-10/drivers/addon/e1000 and then just load the proper "config" to fit the mahcine from the /usr/src/linux-2.4.18-10/configs/ and then edit the Makefile to show a version number without "custom" on the end... you should be able to build the kernel and modules (use make bzImage for best results lately for the kernel) then you'll have to edit /boot/grub/grub.conf to reflect what you just did...
Have fun, if ya want I step ya through it.... *hehehe*
You'll see even better performance, the major version number changed and incremented. From 3.5.19 (I think) to 4.3.2. Made a HUGE change in raw throughput also... got nearly 80% theoretical throughput from this machine... like 5.5GB a minute over GB Enet.
Just a thought... unless it is good enough.
greg - Grand-Master Artist in IT,
curley95@attbi.com -- [link|http://www.iwethey.org/ed_curry/|REMEMBER ED CURRY!!!]
Your friendly Homeland Security Officer reminds:
Hold Thumbprint to Screen for 5 seconds, we'll take the imprint, or
Just continue to type on your keyboard, and we'll just sample your DNA.
|
Post #51,301
9/10/02 9:07:00 PM
|
I knew I was throwing you a bone
Enjoy it.
I forwarded your driver reccomendation
to my sysadmin. I'm sure he's capable.
Thanks.
|
Post #51,305
9/10/02 9:28:23 PM
|
Mmmmmmmmm... Tasty!
greg - Grand-Master Artist in IT,
curley95@attbi.com -- [link|http://www.iwethey.org/ed_curry/|REMEMBER ED CURRY!!!]
Your friendly Homeland Security Officer reminds:
Hold Thumbprint to Screen for 5 seconds, we'll take the imprint, or
Just continue to type on your keyboard, and we'll just sample your DNA.
|
Post #51,453
9/11/02 7:12:45 PM
|
Update
BCM5700 driver was a bit better, but
not much. The default was the "community"
Tigon driver. The BCM5700 is the vendor
supplied one.
SysKonnect:
[link|http://www.syskonnect.com/|http://www.syskonnect.com/]
which rated higher than any card test
in a GB comparative review
[link|http://www.syskonnect.com/syskonnect/performance/gig-over-copper.htm|http://www.syskonne...r-copper.htm]
(yeah, now on vendor site, but not where
I found it originally), sucked just as bad
for NFS writes.
Intel RULEZ.
|
Post #51,460
9/11/02 8:25:23 PM
|
Good to know...
Ever since I actually experienced a REAL-WORLD test of 10/100 cards using the 825** shipsets from Intel and nearly ANYTHING else... I have been... stuck in Intel Gear for E-Net cards.
What I saw was, utilization *AND* throughput were on a relation ship similar to:
x = utilization, y = throughput/workdone
x:y2
I just fell in love... the cards (especially since the E10*+(e100+, e1000+, workstation and server classes) series came out) have never ceased to earn thier keep. I rarely find problems with them... if I do it is usually a bug in the module/driver supplied... and almost always NEWER E10*+ card and/or drivers == better/more stable performance.
Well, take it for what it is worth...
greg - Grand-Master Artist in IT,
curley95@attbi.com -- [link|http://www.iwethey.org/ed_curry/|REMEMBER ED CURRY!!!]
Your friendly Homeland Security Officer reminds:
Hold Thumbprint to Screen for 5 seconds, we'll take the imprint, or
Just continue to type on your keyboard, and we'll just sample your DNA.
|
