WebIntelligence 2.5 and BCA 5.0 Hardware sizing

server administration Web Intelligence 1.x/2.x Servers
1

Hello,

A simple question with, but mayby, not with an simple answer: I have to consult a customer about a sizing issue.

What kind of hardware do I need when a department has 80 WebI 2.5 users who retrieves 10 advanced reports (probably weekly), queried by the BCA.

Secondly, does anyone has a document about hardware sizing issues, with some examples or numbers? Please mail: j_henskens@hotmail.com

(or joost.henskens@roccade-public.nl if its a big doc.)

Please help!
______________________________________________________ Get Your Private, Free Email at http://www.hotmail.com

Listserv Archives (BOB member since 2002-06-25)

2

Joost

Business Objects puts out a rather comprehensive guide on this in the 5.0 release CD. Look on the documentation CD for the “Deployment Guide” which is filename Deployment50EN.PDF (English Version). Chapter 4 is titled “Server sizing, Capacity and Optimization”. Should answer your questions.

Mike McErlain
McErlain Consulting
MikeMcErlain@Netscape.net

At 09:42 AM 3/28/2000 +0000, you wrote:

Hello,

A simple question with, but mayby, not with an simple answer: I have to consult a customer about a sizing issue.

What kind of hardware do I need when a department has 80 WebI 2.5 users who retrieves 10 advanced reports (probably weekly), queried by the BCA.

Secondly, does anyone has a document about hardware sizing issues, with some examples or numbers? Please mail: j_henskens@hotmail.com

(or joost.henskens@roccade-public.nl if its a big doc.)

Please help!

Listserv Archives (BOB member since 2002-06-25)

3

Joost,

There is also a more calculated methodology of sorts that Business Objects presented at the conference. We did load testing here to validate that this was correct and it is. Here it is…good luck! Jodie

All benchmarking was done by Business Objects using a Bi-pro 350 MHz machine on an NT server.

Two Business Objects processes , WIQT and BusObj, determine the server size. A WIQT process is launched for each user when logging into WebI, retrieving the document list, creating/refreshing thin client documents, generating LOVs for full client documents, and processing scheduled thin client documents. A BusObj process is launched for each user when refreshing full client documents and processing scheduled full client documents. The BusObj process takes a lot more CPU due to the added complexity of full client documents.

CPU:
A single CPU supports 12 BusObjs or 30 WIQTs for refreshing documents of average complexity and retrieving 300 rows.

RAM (in MB) for each node:
100 + (50* number of BusObjs supported) + (30*number of WIQTs supported)

Initial Disk Space (in MB) for Cluster Manager: (Number of registered users * 0.4) + (Number of universes *.5) +150

At 09:42 AM 3/28/2000 +0000, you wrote:

Hello,

A simple question with, but mayby, not with an simple answer: I have to consult a customer about a sizing issue.

What kind of hardware do I need when a department has 80 WebI 2.5 users who retrieves 10 advanced reports (probably weekly), queried by the BCA.

Secondly, does anyone has a document about hardware sizing issues, with
some
examples or numbers? Please mail: j_henskens@hotmail.com

(or joost.henskens@roccade-public.nl if its a big doc.)

Please help!

Listserv Archives (BOB member since 2002-06-25)

4

Joost,

There is also a more calculated methodology of sorts that Business Objects presented at the conference. We did load testing here to validate that this was correct and it is. Here it is…good luck! Jodie

All benchmarking was done by Business Objects using a Bi-pro 350 MHz machine on an NT server.

Two Business Objects processes , WIQT and BusObj, determine the server size. A WIQT process is launched for each user when logging into WebI, retrieving the document list, creating/refreshing thin client documents, generating LOVs for full client documents, and processing scheduled thin client documents. A BusObj process is launched for each user when refreshing full client documents and processing scheduled full client documents. The BusObj process takes a lot more CPU due to the added complexity of full client documents.

CPU:
A single CPU supports 12 BusObjs or 30 WIQTs for refreshing documents of average complexity and retrieving 300 rows.

RAM (in MB) for each node:
100 + (50* number of BusObjs supported) + (30*number of WIQTs supported)

Initial Disk Space (in MB) for Cluster Manager: (Number of registered users * 0.4) + (Number of universes *.5) +150

Listserv Archives (BOB member since 2002-06-25)

5

I would like to pose a problem to everyone out there regarding this suggested calculation.

My current configuration is

Master Installation on:
Dual 450 mhz Pentium III - Compaq Proliant 5500 Server 640 MB Ram
30 GB Disk Array

Node Installation on:

Quad 200 mhz Pentium Pro - Compaq Proliant 5500 Server 640 MB Ram
30 GB Disk Array

Both servers are running:

Windows NT 4.0 SP 4.0
IIS v 4.0 (Master Only)
Oracle SQL Net v2.3
WebI v2.5.2
Broadcast Agent v2.5.2

According to the calcualtion I should have:

CPU Requirement: 24 BusObjs or 60 WIQTs Maximum Concurrent User Base ( For this Calculation I will use 10 BusObjs and 20 WIQTs Concurrently)
Ram Requirement: 100+(5010)+(3020) = 660MB Ram Disk Space Requirement: (10*.4)+(25*.5)+150 = 166.5 MB

My current configuration supports these calculations. Now, for the test: I setup a test where I had ten concurrent users logged in and all of them refreshed full client reports at the same time. CPU usage on both server went to 100% and the average RAM usage between both servers was around 200MB.

Two of the ten processes failed to refresh. Both users received COM errors. They had to log off and then back on to continue using the system.

Am I missing something. I know that there are alot of parameters that can be adjusted, however we have had two consultants review our current configuration and (according to them) it is configured for optimum performance.

Are there any ideas out there as to what is going on? This calculation does not hold water for our configuration as it stands now.

I would like to know from others, especially those who have a similiar configuration, to speak up and let us know what they are doing as far as hardware/software solutions to support their respective user base.

Original Message:

There is also a more calculated methodology of sorts that Business Objects presented at the conference. We did load testing here to validate that this was correct and it is. Here it is…good luck! Jodie

CPU:
A single CPU supports 12 BusObjs or 30 WIQTs for refreshing documents of average complexity and retrieving 300 rows.

RAM (in MB) for each node:
100 + (50* number of BusObjs supported) + (30*number of WIQTs supported)

Initial Disk Space (in MB) for Cluster Manager: (Number of registered users * 0.4) + (Number of universes *.5) +150

Listserv Archives (BOB member since 2002-06-25)

6

We did load testing with a similar configuration. Is it spiking to 100% or staying at 100%? Are you isolating the WIQT or BusObj processes on the node? Or are you enabling the WIGenerator and BusObj on both servers? The way we conducted our load testing was to ramp up the logins and then do the concurrent refreshes. The login was part of the intialization in the loadrunner scripts. The refresh was actually what we were testing (60 concurrent refreshes). We certainly didn’t see the CPU on the NT servers maxed for 10 users upon login, but we didn’t have them go directly into a refresh either. How many WIQTs are running at this time? What does the security domain look like when you’re trying this? In other words, what is the %utilization on your unix box?

Feel free to give me a call here at WDW and I we can compare notes.

Jodie
407-566-5177

Listserv Archives (BOB member since 2002-06-25)

7

I have one quick question - are your connections set to wait for resources or to give error? On repository stuff (univ/docs), you would want to wait… that may help on the failures you have.

Brent

Listserv Archives (BOB member since 2002-06-25)

8

What kind of report were you refreshing? How would you rate the complexity? We tested with a simple report on the security domain and limited the query to fetch 300 rows. How long does it take one user to refresh and how much CPU is it consuming for one user?

Listserv Archives (BOB member since 2002-06-25)

9

I would rate the report as simple/moderate. It consists of 19 fields from 8 tables. Currently the report is pulling 984 rows. It is currenly taking 7 seconds in BusObj (Full Client Application) to refresh.

Testing the report on WebI using one user, it takes roughly 15 seconds to refresh. While refreshing it is hitting the server in the 40-60% of CPU usage.

______________________________ Reply Separator _________________________________
Date: 3/31/2000 2:38 PM
What kind of report were you …

Listserv Archives (BOB member since 2002-06-25)

10

The testing BO did and their formulas/methodology were based on an “average” complexity report retrieving 300 rows. So, in order to get their results you would have to limit your results to 300 rows and not do any fancy joins. I wish they had released more details surrounding their tests like response times, utilization %'s, etc… That way we could do baseline tuning of the WebI configurations.

I think the general rule here is to always do a benchmark and use the formulas below for a starting point. Every implementation of WebI will be different in regards to the report complexity. So I’m sure we’ll be load testing/perf tuning for every application that uses WebI here in order to upgrade/add nodes accordingly ( mind you , we have none yet). We will definitely try to use the BCA everywhere we can with regards to WebI due to the system resources WebI consumes. However, I think in doing that it is a sacrifice of the real value of using WebI…ad hoc queries. I am certainly looking forward to any enhancements to WebI that will improve performance and decrease the load on the servers.

What have you seen in regards to the full client refreshes?

Jodie

Listserv Archives (BOB member since 2002-06-25)

11

We are also into similar kind of problem : One of our reports(would rate it as Complex) is working with WebI 2.0 but now not working with WebI 2.5.2 … This SQL generated is having 514 lines of code having several decode ,sum statements .The ratio of Select and where statement is also 1:1 . But whenever we try doing "Run Query " it immediately returns to Web Panel . It is to be noted that the default heap size is same for WebI 2.0 and 2.5.2 ; even increasing the heap size to thrice (as told by tech support…) was of no use .
Pasting the same SQL in BO5 free hand SQL mode leads to its truncation and is of no use . When we are trying to run WebI query thru BO5 its giving an unhandled exception error .
Though if we remove 2 variance objects (out of 7 variance objects which we are having right now ) seems to solve the problem as the same report runs everywhere… but that is no solution indeed … I would really appreciate if somebody suggests anything that can be done about it …
Thanks,
Vikas

Listserv Archives (BOB member since 2002-06-25)

12

Vikas,

Regarding your run query issue…Check your max active time setting for the WIGenerator. It should be set to ((The time it takes to run this query in BusinessObjects)+(The time it takes to Save As HTML)+(The time it takes to open the report in a web browser)).

I believe I also saw this issue with WebI when some of my universe tables weren’t there. Check your universe connection.

Jodie

Listserv Archives (BOB member since 2002-06-25)

13

Jodie ,
I checked WebI generator max active time setting and its set to 3600 sec which used to be 600 sec for our WebI 2.0 on which the same query is running fine …
Regarding the universe connection ,since the same query is running in WebI 2.5.2 when 2 variance objects (out of 7 such objects ) are removed from the query .Is there something that could be changed when i’m removing these 2 objects .All the variance objects are comng from different fields of a single table .
Besides,the same report is not running on any of the shared BusObj Installations (PC Client) …and its running fine on m/c having standalone installation of BusObj 4.0/5.0 …
Is there something else which migh tbe causing this ?

Thanks for your suggestion …
Vikas

Listserv Archives (BOB member since 2002-06-25)