I attended the HP Technology Forum in Houston, TX on Sept. 17-21, 2006, and had a chance to talk with Doug de Werd, HPC Technical Marketing Manager for Hewlett-Packard, about what HP is doing with Windows Compute Cluster Server (CCS) 2003.

Ken: Can you tell us a little bit about what HP has been doing with Microsoft and Windows CCS?

Doug: HP has been working with Microsoft on Windows CCS for several years now. We worked with them very early on to discuss target markets, product features, and so on. Then we started working with their early beta versions in our labs. So we’ve had a really good working relationship with the Microsoft Windows CCS team over the past couple of years.

Ken: What are HP’s product plans for Windows CCS?

Doug: We are supporting Windows CCS in several ways. Of course, we will support customers who get their version of CCS through their Microsoft Volume License relationship, but we are really focusing our efforts on the HP branded OEM version of Windows CCS.

Ken: So what's different about the HP version of Windows CCS?

Doug: Certainly there are things about the core OS and cluster tools from Microsoft that we can’t change, so we’ve focused our efforts on adding value in ease of deployment and performance. The HP OEM version of CCS has the same core OS and cluster software, but what really makes it different is when we use our factory pre-installed image as the basis for deployment. The first thing this does is removes the issue of network topology from deployment. Windows CCS uses Microsoft Remote Installation Services, or RIS, to install the OS and cluster software. RIS has limitations and requirements as to what kind of network configuration you need in order to use it, and we’ve found that this model won’t work for all customers. So by basing our installation on a factory pre-installed image, the network topology becomes a non-issue. The other thing that the factory image lets us do is integrate additional pieces into the image. For instance, we integrate the required Microsoft hotfixes into our image so that customers don’t have to download them from the web and install them themselves. We also incorporate all of the HP drivers needed to support the various ProLiant servers, which ensures that everything is there and the customer doesn’t have to go searching for drivers. We also include InfiniBand drivers for Voltaire IB cards, which we use as the IB interconnect in our Cluster Platform program, and we also include the base code for HP-MPI for Windows, which is the HP version of MPI for high speed interconnects.

Ken: So the factory image is the basis for your deployment. Do you add or modify anything else?

Doug: Yes we do. The other key thing is the factory integration capabilities we have with our Cluster platform and Cluster Platform Express programs. This leverages the factory image of CCS, but takes it a step further. We work with our customers to get cluster specific information, like node names and IP addresses. We then use that to complete the installation on each cluster node in our factory. We’ll also install the Voltaire IB drivers and HP-MPI for Windows if needed. Since it’s hard to integrate directly into the customer’s Active Directory domain, we can’t quite complete the cluster integration in the factory. Once the cluster is at the customer site, there’s just a few simple steps to take to finish it. Basically it is just adding the head node into the domain, and then installing the cluster software on the head node. Once that’s done, you run an HP provided script that adds all of the compute nodes into the domain, and then another HP script to add them into the cluster. The whole idea behind this is that it eliminates the need to manually log into each compute node and perform some action. Theoretically, once the cluster is on site, the customer never has to touch a compute node other than to turn it on.

Ken: That sounds pretty cool, and easy for the customer too. You mentioned focusing on performance – can you elaborate?

Doug: We’re doing several things regarding performance. The first is the fundamental work with the ISVs to make sure that their code runs well on Windows CCS. Our HPC ISV engineering team has a long history of working with the major HPC ISV’s, so were leveraging those relationships for CCS. In some cases we have provided hardware directly to the ISVs and even assisted with porting code. But a lot of our work is done through our lab in Houston. We have several Windows CCS clusters set up that are remotely accessible to both our test engineers and the ISV partners. Our largest cluster is a 64-node, 256 CPU cluster with InfiniBand. We start out testing on the smaller clusters, and once we have the basics down we move them to this bigger cluster for scalability testing and benchmarking. So working with the ISV apps is one thing we’re doing, but the other has to do with improving performance using HP-MPI. We have ported our Linux version of HP-MPI to Windows, so now customers can take advantage of using the same MPI that they have used previously. Many of the ISVs have standardized on HP-MPI for Linux, and now we can offer the same thing for Windows CCS. The bottom line is that we’re helping the ISVs get ready for Windows CCS, so we have the confidence that their apps are going to work great with HP gear.

Ken: Thanks Doug. Any final comments?

Doug: Well, we know that having a good product is not enough to be successful. You have to wrap all the other things around that as well – things like support, services, customer engagement. We have on ongoing relationship with Microsoft to really provide all of the pieces – not just the product – to help Windows CCS succeed. I think there is a real opportunity and pent up demand for a Windows based HPC solution, and we’re putting a lot of effort, along with maximizing our great relationship with Microsoft, to try to bring the best Windows HPC solution to our customers. It was a pleasure talking with you, Ken.

Doug's presentation...