Hello everybody,

We are breaking our head to finalize how many Processors we need for each Windows 2000 Server to run WebSphere V5.

Our goal is "No Single Point of Failure". We need High Availability!.

I have a feeling that instead of 2 cloned JVMS using Windows SMP, I would rather go for 2 physical machines each with 1 JVM only.

How frequently, do we shutdown and restart Machines hosting WebSphere Applications running on Windows compared to AIX and SOLARIS? May be 1 month, 6 months or even less?

I know a large % of the industry goes for UNIX platforms. Please don't ask why we are using Windows? Are there any bottlenecks if we use Windows?

Let's say, there is single JVM in each Windows machine? Will you prefer 4 Processors or 2? is IBM JDK 1.3 going to use ALL Processors in single JVM(called hyper-threading)?

I searched the whole Web, to compare Windows and UNIX w.r.t WAS. Didn't find anything.

I love to share your ideas. Curious to know the industry facing any problems regarding WebSPhere Applications running on Windows 2000.

Thank you.

Definitely run multiple servers to avoid single point of failure. As long as Windows is well configured (good supported hardware and Microsoft-blessed drivers) and doesn't get overly busy (i.e. you keep it running at max 80% utilization at peak times, and generally lower) it will run fine for a couple weeks even up to a couple of months. To avoid SPOFs altogether, you'll need to run on servers with redundant power supplies (one one each of two separate power grids) with two network cards and two networks and two load balancers with a HA bridge. The real problem is guaranteeing no SPOF with your back end (e.g. a database) without spending a small fortune.

Most apps don't need to truly avoid SPOFs ... it's a question of which SPOFs you're willing to accept given the amount of money -- and time -- you're willing to invest.

Peace,

Cameron Purdy
Tangosol, Inc.
Coherence: Easily share live data across a cluster!

Thank you for your nice advise.

Windows usually trashes threads if the process has over a hundred threads or so. Additionally, Windows running in multi-processor mode has a high overhead and tends to use more system time (I've seen upto 25% overhead with CORBA applications) to manage a process' threads.

I would recommend running multiple instances on SMP boxes and setting the affinity of each instance to a single CPU. That way you get more CPU time to run your business and the system tends to be more stable.

As Cameron has pointed out, eliminating SPOFs are directly proportional to how much time and money you plan to spend.

Just my 2 cents to add on the excellent response from Cameron ...... in order to avoid database SPOF, you need to use some high-end db product (e.g. Oracle) and implement features like data replication and failover (e.g. Oracle DataGuard). It is usually recommended to have 1 db as the master and the other(s) as slave(s).

Hello everybody,

Thanks for your nice advise.

By the way, how do you set affinity for Multiple WAS instances in SMP to use a single CPU?

We have high availability solutions for Web Servers. And also from Database point of view, we have Microsoft SQL Server real-time Clustering(defintely lower than Oracle Clustering). We have been running business since long time with the Web Servers and SQL Server. We want to migrate out business logic to WebSphere.

Is it possible to set Processor affinity for each WebSphere instance? I mean, WAS-1 goes to Processor-1, WAS-2 goes to Processor-2 in WIndows SMP? Similarly how do you point both WAS-1 and WAS-2 to a single Processor-1?

Thank you.

By the way, how do you set affinity for Multiple WAS instances in SMP to use a single CPU?

Depends on the OS. You can do it on Windows Server through the task manager. I think it is "pbind" on Unix.

Peace,

Cameron Purdy
Tangosol, Inc.
Coherence: Easily share live data across a cluster!