Windchill 9 on Linux

Here is some documents that might help you out. I've installed PDMLink 9.0 and 9.1 on Linux 4.7 ES and AS at Psion Teklogix and Com Dev International. With the multiple AMD or INTEL quad6 2.6 to 3.16 GHz cores. It has 10X better performance than the HP-UX itaniums. Never had to go to cluster, because the system was performing well with 3 to 5GIG of allocation to every java process including tomcat. The 8 to 16 cores handle all the 5 foreground and 3 background (2 for publishing and 1 for other). All our data is tied to the EVASANXPS. We are now considering updating our network server bandwidth to 10G between the oracle, workers and Windchill. We have over 70 concurrent ProE users with 6 CAD workers on a single workstation. If you have time, I suggest you go with Windchill 9.1 M050 with Red Hat Linux 5.# ES if you just need 2 CPU sockets with unlimited RAM. You can start with a 2X Quad Core and move up to potentially 2X 812 core with 3 channel connection to DDR3 ram which is suppose to be released at the end of this year.

It really all depends on your application and how you plan to use Windchill.

In 2005 at a FlexPDM 7.0 Upgrade Implementation Reebok which is now Adidas in Canton, I found out that with a new free version of java to be able to handle a larger heap memory size of 2 GIG instead of 700 meg,completely removed the PTC requirement of a clustered system. I proved out that the methodservers was not being used and it was solely tomcat to oracle. It could of saved a ton of money of consulting and hardware for this cluster requirement if test were done in advance. I had the original developer of FlexPDM confirm my findings and suggest we cluster Tomcat if requried. There were some additional 4 oracle indexes which were suggested by Oracle Optimizer and toad which was placed into Oracle. These were the old HP-UX 4400s.

In 2008,this is the storey what happened at Psion Teklogix which was a Windchill 8.09.0 and then finally upgraded to 9.1. We bought,

· Production Windchill APP HP-UX Itaniums which were. (oracle guy bought them) Oracle installed on a separate oracle system.
o OS: HP-UX itanium
o dual cores X 4
o max CPU core is 1.6 GHz
o 64 GIGs of RAM
o 4 gig min and 5 gig max for each processor
o <u>http:h20219.www2.hp.comintegrityw1enmid-rangerx7640-specifications.html<u>
· Test HP Intel machines dual cores X4 – Linux 4.7 AS (they were on sale, MIS manager got a deal)
o OS: Linux 4.7 AS
o AMD dual cores X 4 (3.0 GHz core)
o max CPU core is 1.6 GHz
o 64 GIGs of RAM
o 4 gig min and 5 gig max for each processor
o <u>http:h10010.www1.hp.comwwpcusensmWF06a15351-15351-3328412-241644-3328422-3646081.html<u>
· Development machines intel quad cores
o OS: Linux 4.7 AS (kept it the same)
o Intel quad cores X 4 (2.93 GHz core)
o max CPU core is 1.6 GHz
o 32 GIGs of RAM
o 3 gig min and 4 gig max for each processor
o <u>http:h71016.www7.hp.comdstorectoBases.asp?ProductLineId=431&FamilyId=2991&jumpid=re_R2515_storesmProdCatDL300DL380G6<u>
· Our production and test database were on mirrored systems owned by the oracle db which were also the HP-UX itaniums
· All servers were on the same network VLAN and switches with fibre GIG connection
· All servers were configured tuned identically with exception between OS of Linux and HP-UX itanium . Followed PTC performance tuning guide pretty much to the T. Not much on Linux but translated it down.
What transpired was actually funny. Our development training servers which were mirrored of production database and content we starting up in 16 seconds will the production and test were taking over 160 seconds. The same pages and queries were almost instantaneous (1 to 2 seconds) on the development and 2 minutes on production and test. We used Badboy, JMeter, JConsole and top to monitor the systems performance. We literally made our development servers production and returnedrestructured our test and production systems. The performance charts and testing pretty much was in sync with the charts from both AMD and Intel with respect to CPU speed and multiple cores. Limiting the process to within the CPU socket and memory was the huge advantage gain:

<u>http:www.amd.comusproductsserverprocessorssix-core-opteronPagessix-core-opteron-benchmarks.aspx<u>

<u>http:www.intel.compen_USproductsserverprocessorxeon5000technical-documents<u>

At Com Dev Int. Ltd, we are now moving to 2 or 4 X 6 core with AMD or Intel Blades. Increasing the number of method servers to 6 or 8 depending on the number of concurrent users. We are currently on production HP-UX itaniums rp2600 which are maxed out with 2X Dual Core 1.6 GHz CPUs. We are now planning to upgrade to Windchill 9.1 M050 with Red Hat 5.# ES. We have mirrored production WC9.0 systems on Linux 4.8 ES and AS on old simple dell workstations 2X Dual Core CPUs at that is outperforming our production. We are now moving to HP 460c G6 blades with a 10GIGE backbone switch in the chassis so that oracle, windchill and workers are connected at 10GIG for the new upgraded system. We areincreased the number of background methodservers to 3+. We haven’t gotten issues with tomcat yet, with the larger heap size of RAM but we are considering to cluster tomcat. Clustering a system may sometimes cause a slight performance hit to the entiresystem:

• constantly monitoring the cache
• syncing the cache between the master and slaves
• NFS mounts between the SLAVES to the MASTER vault mount. This would cause a slight delay on the master. We use EVA where the master has the single mount to the EVA controlllers. If you use a NAS then this issue would not exist since they all have direct access to the vaults

• We also found out that the backgrounds and foregrounds would get taxed if the publishing was slow. We increased throughput the connection between our worker and server and the Windchill server has a better performance because it doesn’t have to wait for the CAD workers to respond.

  Also, extraxtops on the CAD worker, causes a hit on thequeues of the windchill system. For example, there are constant incidences where you can place 1 to7 CAD workers on the CAD worker workstation (2X quad cores). We have seen even with1CAD worker, there would be hanging xtops with hanging executing queues. At times, the CPU of windchill would constantly get pegged or the reverse opposite

  with no activity and Windchill would grind to a halt. We just have to monitor theCAD workstation(s) very closely to stop all workers andrestart through the CAD agent administrator.

  It’s working amazing and we are saving tons of money of consulting and HW.

  Good luck,

  Patrick