BSFN OCM guide for Citrix

Hi All,

We've been trying to speed up some interactive apps on Citrix by mapping some logic to LOCAL rather than the Ent Svr in the W environments. In many cases there has been large reductions in response times. There lots and lots of CPU intensive non SQL oriented functions that run much better locally.

My question is: Has anyone written up a rough guide listing all the typical candidates, or found a KG document to help with this?

JohnO

I usually map "Minor" BSFN's to local, if I cannot map them all locally. You can use the usual UBE for that with a data selection...

Regards,
Alex.

John,

We tried and abandoned the W environment concept for terminal server years ago. We run business function logic locally on the terminal server, UBE logic on a dedicated enterprise server. Even PeopleSoft/JDE GATS consultants admit that W environments do not enhance terminal server performance.

We do use a dedicated enterprise server for processing business function logic. That server supports our web based clients and our XPI servers.

Our enterprise server configuration:

1) Two eight-way Compaq 8500 servers clustered together to run the database and serve as the primary enterprise server. The servers are active/active with SQL running on both nodes. On node one we have the Production and system databases and JDE. On node two we have SQL running the PY and DV databases.

2) A dedicated enterprise server for processing UBEs and acting as the security (login) server.

3) A dedicated enterprise server for processing BSFNs for the web-based clients and the XPI servers.

4) Client Side - 18 Citrix terminal servers and one Websphere server giving the users the option of a web or citrix interface.

Gregg Larkin
Praxair, Inc.
North American PeopleSoft
Enterprise One System Administrator

We map all our BSFN's local for our Citrix servers (40 in all). There are a few BSFN's that we have to run on the enterprise server for things like Vertex, etc. In general I think the W environments were based on the concept that your Citrix servers would be under powered compared to your enterprise server and thus would need this effort off loaded. With todays high power Citrix servers this isn't much of an issue for most businesses. Here are the ones that we have mapped to our enterprise server....

Object Name
B0000182
B0000183
B3800070
B7300002
B7300004
B7300012
N5541001
N5542007
N5542008
X00TAX

John,

I know of no PeopleSoft (JDE) document, but we found the effort to come up with our own VERY effective. We use most modules (Finance, Mfg, OE & Dist, Fixed Assetts, etc) but not HR. We also use advanced pricing and advanced scheduling. Most of our heavy hitters (BSFN's) move a large amount of data. By mapping the MBF's to the enterprise server we dramatically improved response time, for the simple reason that no matter how fast your LAN it takes a long time to move 1G of data. Our best example was that Supply & Demand was taking 53 min to return with the info, why they have their rules set like the do is a differnet matter, when we moved the BSFN's to the ES, this dropped to 3 min.

We saw less dramatic improvement from OE where they use advanced pricing (avg 10000 recs per item) and advanced transportation (we go down to the truck at OE time). On a 100 line order we improved our calc time from minutes to seconds.

Bottom line, if you can get by with mapping everything locally, that would be my PREFERRED choice. If this just isn't feasable, bite the bullet and come up with your own.

Tom Davidson

John,

There is one piece of information that Tom Davidson left out, his company is running some REALLY BIG HONKING enterprise servers. A 12 processor AS400 is one big bad computer and will kick the crap out of any Citrix server configuration that you could come up with. I'm not an AS400 guy but I will admit to having server envy!

As far as the vertex mappings mentioned previously (reminder: B0000182 B0000183 B7300002 B7300004 B7300012 X00TAX), we fiddled around with mapping that local and on the enterprise server. The configuartion that worked out best for us was mapping the business functions locally. The catch was that we needed to install the vertex utility software on all of the enterprise servers, all of the citrix servers, and on all developer fat client computers that did any development that touched vertex. The vertex software is pretty easy to install, so that worked out best.

Gregg Larkin
Praxair, Inc.
North American PeopleSoft
Enterprise One System Administrator

Gregg,

Just a note. While it is true that we NOW have a 12-way. The results I posted were from a 4-way box. Also while we have a 12-way I currently have 9 of the processors turned on. I currently have about 200 users per cpu, I am also currently running 67k (67,000) UBE's through the box per week. This is an average, my peak at month end is actually around 85K UBE's for the week.

Just a follow-up.

Tom Davidson

Hey thanks to everybody for those replies - most helpful.

Cheers,
JohnO

Gregg,

We are also running approximately 28 Citrix servers with MetaFrame XP and we are considering dumping the W environment in production Xe before going live with 8.9 this summer. How many concurrent users do you have (Citrix only) and how many users per Citrix server during peak times? This is some timely information considering my CTO is asking me for proof that other customers are doing this without the need to double the number of Citrix servers.

We are going to run performance measurements on our TSE's and App servers on a heave volume workday, and repeat the process the following week with everything running locally (i.e. PD7333 environment).

Any info you can provide would be great.

Thanks,

Charles,

I have over 3000 potential users. My citrix farm is broken into several different pools. My primary production pool when at full load sees about 35 concurrent users per terminal server. Those users use a backend JDE server for processing UBEs but use the CPUs of the terminal servers for processing BSFNs. The terminal servers are two-way Compaq DL360s. I have several J environments to support the web. The J environments have their BSFNs mapped to a four-way JDE application server and their UBEs mapped to the aforementioned batch server. When I log into JPD7333 on a terminal server, I notice a considerable lag time since the application has to reach across the network for processing the BSFNs. The 4-way application server is not heavily utilized, yet J environments on terminal server are still significantly less efficient. I think your CTO will be pleased with the performance increase in a simple remapping of default BSFN.

Gregg Larkin
Praxair, Inc.
North American PeopleSoft
Enterprise One System Administrator

Gregg,

Thanks for the quick response. This helps out tremendously. I've already composed a rambling dissertation to management. We have a similar hardware setup for the Citrix farm but our pool of users is larger. It's currently sized for 1200 concurrent users, but never seems to get much higher than 850. I take it all of your servers are configured with more than 2GB of RAM? We just went live with Solution Explorer in PD as prep work for the 8.9 upgrade. Memory utilization seems a little bit higher than it was before. To compound the issue, the published app is configured for 64K colors. It used to be 16 and then 256, but some of the web stuff like our Intranet looks absolutely terrible (and performs bad to boot) with anything lower than 16-bit color. I've been telling people for the past year that we should move away from the W environments, but it's hard to convince folks to change the architecture when most of my "evidence" is based on conjecture from this list. I think there are some really bright people posting here and I trust you as one of those folks.

Thanks again.