Performance issues

[sobak]

I have a Netware 6 box running SP3 with the following Config.

Dell 2450
1 Gig Processor
1024 Megs Memory
5 x 36 Gig HDD
1 Gig Ethernet Card
Perc 2 RAID Card

The problem that I am having is that I often receive spikes to 100% utilization that can stay there for about 10-15 minutes. At that time my dirty cache buffers sky rocket and users are getting locked out of their CAD files.

I've looked at all the information, I suspect it could be memory related but I wanted to be sure. My cache buffers (dividing total cache buffers by original cache buffers) puts me at 62%. Novell states anything under 40% should be addressed by memory. My Long Term Cache Hits are at 78% I believe (if I'm not mistaken) Novell recommends anything over 85%.

My thought on this is memory but I want to rule out a possibility with the Disk sub-system as well. The server has been pretty reliable so far, no hardware failures in the 2 years it's been in service.

Anyone that has thoughts on this matter would be appreciated.




david e
*end users are just like computers, some you can work with...others just need a simple reBOOTing to fix their problems.*

 

[terry712]

is it using nss ?

what's you lru at?
does you service processes increase and what are they

you probably do need some more ram but what has changed since this issue - more users , bigger files?

whats you disk space like - do you have large purgeable areas

during this peak - whats the busiest threads
is it ds related ?

 

[sobak]

Terry,

No NSS, we're running traditional volumes.

* LRU's are in question, I just checked them and I'm sitting at 13:46, they've been as low as 5:32.

* Service processes stay the same, I'm currently sitting at 109

* The only recent change that I can think of is heavy archiving, we've moved about 15 gigs of files off the server and onto DVD's. Other than that nothing has changed on the server either hardware or software...

* On the busiest thread issue, the only thread that I can see is a server.nlm waiting for work. Most of the time that will be the largest percentage of utilization on the Imanager view during 100% utilization. Also I get hits on my Cache monitor during this time that leads me to believe this is more memory related than hard disk subsystem.

I have an extra 512 megs that I'm going to throw into it to see if I can get the LRU's back up a little if it does help a bit I think I'm going to MAX the server out at 2 Gigs.

I was worried about this because the file server has been running fine for about 8 months now out of the blue we started to have problems with it. We rebooted it during lunch today to start everything out fresh, funny how it didn't help much all the figures that I gave you were about 45 minutes after the reboot. We currently have 160 connections to the server but we've had up to 300 connections and it didn't miss a beat, this is what has me a little confused. As far as someone adding something or changing the server config, I'm the only one that knows how to do something like that, plus I'm the only one with the passwords to do that, I've been down that road before and didn't want to revist that again.

Let me know what you think....



david e
*end users are just like computers, some you can work with...others just need a simple reBOOTing to fix their problems.*

 

[terry712]

well the fact that its traditional volumes means that this isnt masking the figures (would consider nss though - always suspicious of it on 5/5.1 but on 6 - it's solid)

lru definately points to the slap more ram theory

would definately check your purgeable space and the amount of directory entry - as a large amount of purgeable space or full dets can cause this

 

[sobak]

Terry,

Thanks for your points, I've added another 512 stick in it this morning, going to see if that fixes the issue. I'll let you know.

As far as NSS I wanted to install it when we upgraded but the problem I ran into was that NSS didn't allow for file compression, I'm running it on my server now and to expand all the data to it's native size would push my volumes past the physical limit. I'm hoping during the next wave of hardware refreshes I'll be able to do that but right now I archive off about as fast as they put it back on. I guess it's nice to know we're doing more work to keep my mortgage company happy

Anyway I'll let you know how today goes, after looking at the LRU's I'm sure I'm on the right track.



david e
*end users are just like computers, some you can work with...others just need a simple reBOOTing to fix their problems.*

 

[Ed Andres]

There are some post SP3 patches and settings that are being recommended for high utilization by most SYS OPS. If you have not already applied these, I would reccommend that you do.

nw6nss3c (Even if you are not using NSS, _Admin is NSS)
tcp607jrev2
nls603ft
Go into monitor - server parameters - NCP, and set client file caching enabled = OFF
and Set level 2 oplocks enabled=Off;
run purge_nw.exe once.

They have solved some utilization and back-up issues I was having.

HTH
Ed

 

[sobak]

Ed,

Great I'll do those sometime this weekend and let you know how it turns out. Its early in the day and I've had one Cache performance issue already but my LRU's look like they are holding steady at 28:##. With an occasional drop down to 18 but they are still above the 15 mark. I'll keep tweaking with it and let you guys know where I stand.

Thanks



david e
*end users are just like computers, some you can work with...others just need a simple reBOOTing to fix their problems.*

 

[sobak]

A little more interesting information.....

I didn't apply any of the patches Ed has outlined, waiting for this weekend but I did notice something of interest.

This server was originally 4.11 and we did an inplace upgrade to 6.0. Under 4.11 I had my Directory Cache buffers set to 1000 min, 3000 max. After my memory upgrade and restart yesterday my setting stayed at 1000 and only increased a little. This prompted me to do a little research on this.

The official Novell explanation is that when a server starts up with the Min cache buffers set to 1000 and it only needs 600 it will only take 600. At the time when the system needs more cache buffers, NetWare will take what it wants up to the Minimum. Here's the tricky part, once the system hits the Minimum setting then when additional buffers are needed then it has to wait for the Directory Cache Allocation Wait Time to expire then it will take more.

Armed with this information I set my wait time to 1 second and monitored it over a 1 hour period. The normals problems were still occuring, excessive dirty cache buffers, high utilization and people being locked out of files.

After the 1 hour period expired then I went ahead and increased the min cache buffers to 2000, immediately it started to climb until it reached 2000. I set it to 2500 and it took all of those. I left early yesterday but before I left I took the setting up to 3000 and waited until it took all of those. After 4 hours of continued use yesterday (after I left) I called to see how it was working, they told me that it seemed to settle down a little and operate a little better.

Now my settings were fine for my server under 4.11 but I'm wondering if they are inadequate for 6.0. Should I increase the min setting higher so that it starts out with a higher cache setting? My thoughts were that if Novell needed them it would allocate them as needed, I just don't see that it's allocating them as fast as the system is requiring them.

Any thoughts from you guys???

david e
*end users are just like computers, some you can work with...others just need a simple reBOOTing to fix their problems.*