Actually one hot topic in the business is the use of 64 Bit Operating Systems and x64 programs.
Since the release of IBM Cognos 8.4 BI Customers and partners of IBM may also choose between a 32 bit or 64 bit version of their product.
But what are the differences and the benefits of those?
Since not all customers are fully aware about the differences I decided to do some tests combined with giving some more detailed information about the product.
First of all, some of you may be dissapointed because its not the entire product which is 64 bit, just it´s application engine (Web Interface also known as Cognos Connection) runs in a 64 bit application server. The report validation and processing engine (which is written in c++) still runs in a 32 bit mode.
This means that currently "only" web requests in a 64 Bit IBM Cognos environment can be handled faster and more efficient through the 64 bit java application server. Surely the overall main logic may benefit from the greater adress space and faster handling of requests but forget the idea to install a 64 Bit IBM Cognos 8.4 Server to get rid of some long running reports without changes on database side.
But let´s go to the details :-)
The environment:
32 Bit Version
Intel Core 2 Duo processor with 2.4 GHz
4 GB of RAM
Windows 2008 Server Standard Edition (32 Bit)
SQL Server 2005 Developer Edition with SP3 (32 Bit)
Cognos 8.4 (32 Bit)
64 Bit Version
same Hardware
Windows 2008 Server Standard Edition (64 Bit)
SQL Server 2005 Developer Edition with SP3 (64 Bit)
Cognos 8.4 (64 Bit)
The test:
3 new Report Studio reports based on MSAS 2005 and the Cognos sample package GOSalesFact_EN_MSAS20052 new Report Studio reports based on SQL Server 2005 and the Cognos sample package GO Data Warehouse (query)were run in at least a minimum of 70 to 100 iterations. The fastest report ran in 37 seconds, the most time consuming report completed after 9 minutes for one execution.All values are based on averages given from the No. of iterations.
To get an reproducable test case by avoiding any caching or other side effects I decided to use the rstest utiltiy which inherits the main logic of report processing. This tool is often used internally by IBM Cognos Information Management people and has some interesting features. You can find this tool in the Cognos 8.4 bin directory.
The results of rstest may not be comparable to an execution within Cognos Connection, since the bibustkservermain process caches a lot of data. In other words the test results should reflect a "first time execution in Cognos Connection" and therefore the maximum processing time a report may need. As already pointed out, the rstest and bibustkserver processes are both 32 bit programs.
Looking at the result table above, you may raise the question why the elapsed time for the 64 bit tests were always lower then the 32 bit values?
Answer: The database server and olap server (both 64 bit) were able to return the results faster back to the IBM Cognos report engine, this caused a faster execution.
Since the release of IBM Cognos 8.4 BI Customers and partners of IBM may also choose between a 32 bit or 64 bit version of their product.
But what are the differences and the benefits of those?
Since not all customers are fully aware about the differences I decided to do some tests combined with giving some more detailed information about the product.
First of all, some of you may be dissapointed because its not the entire product which is 64 bit, just it´s application engine (Web Interface also known as Cognos Connection) runs in a 64 bit application server. The report validation and processing engine (which is written in c++) still runs in a 32 bit mode.
This means that currently "only" web requests in a 64 Bit IBM Cognos environment can be handled faster and more efficient through the 64 bit java application server. Surely the overall main logic may benefit from the greater adress space and faster handling of requests but forget the idea to install a 64 Bit IBM Cognos 8.4 Server to get rid of some long running reports without changes on database side.
But let´s go to the details :-)
The environment:
32 Bit Version
Intel Core 2 Duo processor with 2.4 GHz
4 GB of RAM
Windows 2008 Server Standard Edition (32 Bit)
SQL Server 2005 Developer Edition with SP3 (32 Bit)
Cognos 8.4 (32 Bit)
64 Bit Version
same Hardware
Windows 2008 Server Standard Edition (64 Bit)
SQL Server 2005 Developer Edition with SP3 (64 Bit)
Cognos 8.4 (64 Bit)
The test:
3 new Report Studio reports based on MSAS 2005 and the Cognos sample package GOSalesFact_EN_MSAS20052 new Report Studio reports based on SQL Server 2005 and the Cognos sample package GO Data Warehouse (query)were run in at least a minimum of 70 to 100 iterations. The fastest report ran in 37 seconds, the most time consuming report completed after 9 minutes for one execution.All values are based on averages given from the No. of iterations.
To get an reproducable test case by avoiding any caching or other side effects I decided to use the rstest utiltiy which inherits the main logic of report processing. This tool is often used internally by IBM Cognos Information Management people and has some interesting features. You can find this tool in the Cognos 8.4 bin directory.
The results of rstest may not be comparable to an execution within Cognos Connection, since the bibustkservermain process caches a lot of data. In other words the test results should reflect a "first time execution in Cognos Connection" and therefore the maximum processing time a report may need. As already pointed out, the rstest and bibustkserver processes are both 32 bit programs.
Looking at the result table above, you may raise the question why the elapsed time for the 64 bit tests were always lower then the 32 bit values?
Answer: The database server and olap server (both 64 bit) were able to return the results faster back to the IBM Cognos report engine, this caused a faster execution.
The last column which is calculated from the rstest cpu time vs total elapsed shows clearly that the time the IBM Cognos process needed to finish was even higher than on the 32 bit system.
One reason for that could be the emulation for the 32 bit processes on a 64 bit system. Those emulations cost surely some time.
Conclusion:
64 Bit is not necessarily faster than 32 bit. The 64 Bit version of Cognos 8.4 BI may improve performance for Cognos Connection activitites, especially if you have a high number of web users which are heavily using Cognos Connection. But don´t expect that your reports will run faster because of the installation of a 64 Bit version of IBM Cognos 8.4 BI.
Hi Sebastian,
AntwortenLöschenI am about to start a project with Cognos 8.4 BI, on:
Windows Server 2008 64bits
Sybase Adaptive Server Enterprise 15.0 SBE Winx64
So I guess we need Cognos 8.4 64 bits version.
Do you know if all components are available for that version? i.e.: Framework Manager, Transformer, all the Studios, etc?
In case we choose the 32 bits version, any considerations we should take?
Thank you very much,
Ana Z
Don´t worry. The 64 bit version contains all neccessary components.
AntwortenLöschen(Even if they are x32 and not x64)
Hv fun!
Hi Sebastian,
AntwortenLöschenplease have you any guide or indication about rules to follo for the server capacity planning oy Cognos 8.4 ?
Thanks in advance, Mark
Hi Sebastian,
AntwortenLöschenWe are moving our SQL Server Database and content store to a seperate server. We are planning on this server to be 64 bit. If we have our current Cognos BI Server on 32 bit will it cause any issues connecting to the Content store on the new server even though it is running on 64 bit?
Thanks for you repsonse.
Greg
Hi Greg,
AntwortenLöschenShort answer.
This will be no problem. :-)
Regards,
Sebastian
So is your conclusion that 64 bit is faster than 32 even if not all components of Cognos are 64?
AntwortenLöschenWhat if the database is 32 and Metric Studio/Planning 8.4.1 are used regularly is 64 still worth it?
Thanks,
Marie
Sorry, I missed seeing your actual conclusion since I came from another site that was a repeat of your blog minus the conclusion. I see now that you state the performance as not necessarily faster except on the front end (Cognos Connection). I wonder if that is worth my trouble considering the DB is 32 bit SQL 2005 and the Planning and Metric Studio are heavily used.
AntwortenLöschenThanks Sebastian. This is such a big help for me. We are given a new VM and Cognos 8.4.1 both 64 bit and 32bit. I was always thinking that if its 64bit, it would definitely be faster in processing my 500 + reports for 350 users. But the catch is I wasn't able to use the Access Manager with this. So, I had to go back to 32 bit so that I can configure the CM with Access Manager.
AntwortenLöschenMy question is:
Given a condition to use the access manager, can I install 64bit and 32 bit and configure 32bit to use Access Manager and 64bit for the rest. ?
You will be able to mix up the processor technologies with a distributed installation. (Separate Web GW 64 bit, sep Content Manager 32 bit) But neither I nor IBM Support would recommend this, since there are risks by doing this.
AntwortenLöschenThe report processing itself won´t be faster by doing this as stated in my post.
In your case you will have to install content manager (the dispatcher) in 32 bit due to your requirements.
But unfornalety even this content manager (dispatcher - java process) would be the most important component (with the most benefit) in a 64 bit installation scenario.
In conclusion:
Install everything in 32 bit and good luck.