Server clustering made simple

HP bundles ProLiant servers with PolyServe’s robust clustered apps for maximum flexibility

Getting top performance from a storage system and making applications faster and more resilient are two critical and challenging aspects of computing. Unfortunately, many vendors’ solutions focus on just one side of the problem or require massive, costly infrastructure changes.

If that scenario sounds all too familiar, consider PolyServe Matrix Server, which combines a robust cluster file system with a virtual server architecture that enables fail-over and flexible allocation of computing resources. Matrix Server doesn’t require proprietary hardware and it runs on almost any Linux or Windows-compatible server or storage device, so it’s easy to see why PolyServe captures the attention of partners such as HP, Microsoft, and Novell.

HP recently began bundling PolyServe Matrix Server with its StorageWorks Enterprise File Services Clustered Gateway . I looked at Matrix Server bundled with HP’s minimum configuration of two ProLiant DL380 servers running Windows Storage Server 2003. It’s a natural fit; HP’s servers are a solid foundation for clustering, and the PolyServe software is the star of the show.

Squaring Off with the SAN

Imagine a grid of applications and virtual servers that an administrator can easily and dynamically interconnect from a management console to provide fail-proof or load-balancing configurations. That’s PolyServe Matrix Server in a nutshell.

Explaining how Matrix Server works, however, is much easier  than installing it, so purchasing the HP StorageWorks bundle with the applications already loaded gives you the benefit of having the product preinstalled on HP’s servers and the option of ordering installation services.

Using Matrix Server, you can create a cluster of as many as 16 Linux or Windows servers and dynamically assign multiple nodes to support your application. To allocate resources from those nodes to your apps, you define virtual servers and assign one or more to each application.

Another key component of Matrix Server is PSFS (PolyServe File System), which allows all the nodes in a cluster to safely share access to the same SAN volumes. Therefore, you can summon multiple servers to speed up your application and to provide immediate fail-over.

Activating these capabilities requires a rather complicated setup that touches just about every device in your datacenter, including servers and SAN devices. For example, to monitor the status of the virtual servers, you need to connect each managed node to a service network over which the Matrix Server will constantly poll the status of each machine.

You control servers’ access to your SAN via usual zoning techniques, but Matrix Server will automatically discover which LUNs (logical unit numbers) are available and monitor their status by following the IP address of your FC (Fibre Channel) switches. Also, your SAN volumes need to be imported into the Matrix Server managed pool and equipped with PSFS.

Many settings are human-driven, so errors are possible, but PolyServe includes very good documentation as well as an application to check that the major installation requirements are met. In the process, it creates a handy status report of errors and successful settings.

Once set up, I opened the Java-based Matrix Server management console. The console has a grid with a row for each application and a column for each virtual server. In my test bed, those apps were Windows file shares, but PolyServe also controls databases and Web servers on Linux or Windows.

The console’s intuitive color-coded icons show whether the applications on particular rows are supported by servers on particular columns. Determining the number of virtual servers behind each app is as easy as counting how many green icons line up on its row.

Getting on the Grid

You can manage other administrative tasks from the same GUI. From a context menu, you can dynamically remove or add servers according to each application’s performance demands. I also added another shared folder and assigned virtual servers to support it with the help of PolyServe’s streamlined wizards. The wizards collect data such as the name of the share, its physical path, and the number of users allowed in a single window for easy input.

I chose how many virtual servers to assign to an app from a box that lists all physical nodes in the cluster. I selected two virtual servers and saw the change immediately reflected on the main screen matrix as two green arrows lined up with my new file share. That’s all you need do to set up a redundant configuration, and that simplicity more than compensates for the time spent on a rather tedious installation.

From a user’s perspective, using a Matrix Server share is exactly the same as working with any other shared folder. To test my new share’s resiliency, I started up a movie on my client then unplugged the LAN cable from one of the two nodes assigned to my share. The movie continued undisturbed, and one of the console icons associated with my share turned from green to red, indicating a failed server.

After I restored the connection, the icon returned to green -- and my movie was still running. I only had two servers assigned to that app, but Matrix Server’s clustered structure can provide multiple layers of fail-over.

Next, I tackled the task of adding a new SAN volume to my cluster and creating a new share. Matrix Server won’t help with the initial steps of creating a LUN and adding it to the zoning; you’ll have to use the native tools of your array, switch, or other datacenter devices.

I clicked “Storage” then “Import disk” on the console menu and selected my new LUN from a list. After a successful import, I created a file system with one mouse click. In just a few minutes -- and without service interruption -- I greatly increased the storage space available to my cluster.

Attaching more servers to the cluster is equally easy and adds processing power that can be shared among all users. PolyServe’s documentation says that each additional server increases performance linearly, but I wasn’t able to reproduce that condition, because my clients didn’t have enough processing muscle to saturate even one server.

A Good Match

I see many interesting products, but the combination of Hewlett-Packard’s clustered hardware and PolyServe Matrix Server offers a unique combination of tools that can pull out the fastest performance and superior reliability from your SAN as well as your  servers. Installation may be a bit of a challenge and deployment could require additional equipment. Those two difficulties, however, are problems that HP’s bundled offering easily solves -- for a fee.

InfoWorld Scorecard
Ease of use (15.0%)
Performance (20.0%)
Management (20.0%)
Scalability (20.0%)
Value (10.0%)
Interoperability (15.0%)
Overall Score (100%)
StorageWorks Enterprise File Services Clustered Gateway with PolyServe Matrix Server 8.0 9.0 8.0 9.0 8.0 9.0 8.6