Performance - General

[froncement]
We plan to have a OneWorld platform with about 1200 users upon MSSQLServer2K / MS WIN2K :
- several Citrix servers (one serving up to 50 users)
- several batch servers
- but only one SQLServer with many databases.
At the moment, about 200 users are working on this platform live.
Does anybody have such a platform and how about performance, system problems,....?
Than you.

Patrick :

If I were you, wouldn't put 1200 users on an W2K/SQL2K Xe installation.
According to what I've seen (and suffered), 250 to 400 users would be
the upper limit for that platform.
Feel free to email for further comments,

Sebastian Sajaroff
B7321 to Xe, NT/W2K, SQL6.5 to 2000, Citrix, JAS

Bonne chance avec OneWorld!

With Standard versions of W2K and SQL2K - maybe, but with Advanced (aka
Enterprise) versions it's still a long way from the upper limit...

Of course, with more than 500 users they will need to start treating SQL
more seriously - the same way as Oracle users usually treat their servers:
it will take a full-time DBA or two to make it perform well and the hardware
cost will be comparable to that of an RS/6000.

Regards,
Alexander Pastuhov.

Patrick,

Because OneWorld is usually CPU-intensive and SQL is (ultimately) -
disk-intensive, the two can live happily on the same box.

By separating SQL from OneWorld you are introducing the obvious network
performance issues into the picture and, generally, you can expect the same
jobs to take twice as long to run (in average) on a separate Logic server as
opposed to running them on the same server as your DB.

As a matter of my personal preference, I usually recommend installing the
two on the same box. The catch is to size the box accordingly...

Re: RE: Performance - General

"Because OneWorld is usually CPU-intensive and SQL is (ultimately) -
disk-intensive, the two can live happily on the same box."

Sorry but I am feeling crotchety today and someone gets my wrath.

OneWorld and SQL are both *memory* intensive and really love it when you seperate them onto two boxen.

"you can expect the same jobs to take twice as long to run (in average) on a separate Logic server as opposed to running them on the same server as your DB."

Without data to back it up, your statement is pure speculation, and wrong speculation at that. Do you have test results to back up this statement?

Re: RE: Performance - General

Wow Brother,

Get a cofee and you will feel better !

Brother, What will be "your sugestion" to this post then. You are saying what is wrong but you don't say what you think is right !

Your input are welcome !

Christian Audet

Re: RE: Performance - General

If you look closely you will notice that my suggestion of separating the two is a part of my last post. Also, over the course of the last two or three years, I have shared on this forum my experiences of what works and what doesn't. If you have a sincere desire to make yourself aware of my humble opinions please avail yourself of the search feature.

Look, I don't want to get into it here but you must agree that on a forum as respected as JDEList it is imperative that information and advice be challenged and reviewed by peers. Sometimes this occurs in a polite, genteel manner and sometimes it doesn't. I sometimes challenge the validity of postings here just as I am sometimes challenged. I think that it is very important that correct information be posted and peer review is one of the ways to assure that occurs. It ain't always pretty but it always results in if not correct, at least varied opinions.

RE: RE: Performance - General

Sure, there is no need to get personal, just professional: I have seen it
dozens of times and measured the timings myself: A simple AB report takes 1
min when executed on a very busy Enterprise server (550MHz zeons) with the
DB on it and 2 min when executed on an idle secondary server (1000MHz zeons)
- 100Mb fully switched EtherNet.

With Oracle or AS/400 this difference goes up from 2 times to 3 times.

Another example that springs to mind is R07200 that takes 20 min on the
AS/400 where the data is and ~1h on a (faster) secondary server.

As is clear from my example, the CPU speed has little to do with it since
the network layer becomes the bottleneck for those UBE's that read a lot of
data as opposed to those that do a lot of calculations.

For reports that use worktables, this is likely to become 4-5 times for SQL
and maybe up to 6-7 times for Oracle and AS/400.

I would welcome any comments (with actual timings) from the users who can
test it...

As for the question of what uses what, the memory is quite irrelevant unless
it is really scarce. On a well sized and configured system there is no
contention for memory at all, only contention for CPU (mainly UBE's), disk
(mainly DB) and perhaps network in some cases.

Patrick,

We are running the environment that you just described, and then some. My enterprise server is a clustered 8 way server with 3 Terabytes of diskspace and 4 Gb of RAM. We are running 9 production terminal servers, with seven more coming on-line as my user count continues to grow. SQL is my primary consumer of CPU resources, but in general our CPU usage hovers around 30%. Our memory usage tends to stick around the 3 Gb mark.

Performance issues - we went live with only 4 CPUs and found that SQL was choking our performance. Our initial configuration of our terminal servers and deployment server was a bit shy of disk space, but we bumped them up in size. We are using one batch server for production and a second for testing.

We are currently using Active/Passive clustering but are looking at several different Active/Active scenarios to make use of all the hardware. We'll be testing several different scenarios on our sandbox servers to determine the optimal configuration. We will test and benchmark Active/Passive, Active/Active, Active/Active with multiple SQL instances.

Gregg Larkin
Praxair
Win2K, Sql2K, XE, SP 19, Update 6

3 terabytes!!!! Are you running General Motors' financials as well??

How many users are working live at the moment ?

Re: RE: Performance - General

Thanks you Brother,

I appreciate and understand your comment, sorry if I got too personal in my last reply.

Maybe I was expecting too much of a answer to some of the performance problem I also have here. I'm into performance Issue also and I don't have a clue of the answer (I've tried almost everything). To give you the picture I have conversion that I ran last year at a transert rate of 4000 records per minute (LOCAL) and this year the same conversion have a rate of 900 records per minutes (ON THE SERVER) and it's worse on LOCAL. INPUT AND OUTPUT DB plus process was tested on the same machine. Now I'm testing input and output on server 1, process on server 2 and it seems to give the same result.

Ok, that's the picture and that's also why I was seeking answer here and in the archive.

Christian Audet

Simon,

You asked "3 terabytes!!!! Are you running General Motors' financials as well?? ". You know what they say, A terabyte here, a terabyte there, pretty soon adds up to some real disk space.

Gregg

Re: RE: Performance - General

Christian,

transert rate of 4000 records per minute
minutes (ON THE SERVER) and it's worse on

Have your databases grown significantly? I have found that many DB's do a
good job on small tables, but after they expand past some 'magic' size then
they need to start a significant number of paging operations. This is
especially

OW 7333 SP 19.1, WIN2K, TSE 2000, Metrframe XP FR 1, CO SQL 7.0

Re: RE: Performance - General

Christian:

I have been in the same situation as you are. In my case, the slow behavior was due to the time the database was needing to allocate hard drive space for the database itself as well the logs, specially if the database grows big.

I will recommend you to pre-allocate the space before starting with the conversion.

Regards
Jorge

Patrick,

You asked how many users we have live in Production. We have 1550 current users. 130 from Canada, 620 from Mexico, 800 from the US. We have another 700 or so US users coming on board by Mid-November. And yes skeptics out there, that's running on Wintel.

Gregg Larkin
Praxair Global CNC
XE, Win2k, SQL 2K

Re: RE: Performance - General

I am currently looking after 3 production instances (Europe, USA & Australia) and my predecessor had the following set up:
Oracle on Win2K server also running JDEdwards logic with a bank of Citrix TSE;s underneath.

Unfortunately our Director will not let us change to Unix or similar to run Oracle on, so this is how I have set it up:

Oracle running on Win2K 8 GB Ram & 8 Processors (No JDEdwards install)....
Four logic servers running JDEdwards with Citrix servers underneath.

The speed is pretty good and the reliability is high - we have the four servers running as security servers and OCM mapping set so we can change over to different logic servers if one should fail. Obviously we are running Fibre between these boxes which makes a big difference. We have around 700 consecutive users during the day in the USA and have no issues.