Veritas cluster roadmap raises the ceiling
256 mixed nodes and beyond
Veritas Vision Veritas has mapped out a bold future for its clustering products, including a significant change in the way future clustering software will be sold and in the ways it can be used.
The Register managed to find its way into a clustering session here at the Veritas Vision conference in which two executives plotted out the future of Veritas Cluster Server. If you are a current or prospective user of the software, the information provided during the session could well be key to your future clustering plans.
One of the major revelations made by Veritas is the decision to bifurcate its clustering line into two separate products. In mid-2005, Veritas plans to introduce a product currently named Next Generation Cluster. This software will be more of a product suite rather than a single software package. Veritas plans to take its current Cluster Server software and then add in bits of its OpForce provisioning technology and Ejasent performance tuning technology. Customers should be able to move software from server to server in a much more fluid way than in the past. In addition, it could help customers increase the utilization of their servers - which helps lower hardware costs - while still maintaining a high-levels of availability and performance.
But here's a bit of potentially bad news for current Cluster Server customers. The OpForce and Ejasent add-ons won't be bundled into future releases of Cluster Server. Veritas intends to sell the Next Generation Cluster suite on its own and then sell OpForce and Ejasent "modules" to current Cluster Server customers. So, if you were hoping for a cheap all-inclusive upgrade, tough luck.
On the positive side of things, Next Generation Cluster should give users some of the most sophisticated clustering tools around. The product is specifically designed to tempt customers moving from two server clusters to much more complex multi-node builds. For example, Veritas Cluster Server today only supports up to 32 connected servers in a single cluster domain. When Next Generation Cluster arrives in 2005, Veritas intends to support up to 256 servers in a single domain, said Jim Senicka, a Veritas product manager.
To support this many systems Veritas is re-engineering its policy engine that helps maintain the coherency of data in a cluster. Instead of distributing this engine across numerous servers as it currently does, Veritas plans to centralize the policy engine. A smallish cluster, for example, would run the engine on two servers. The policy engine, however, would still be running alongside other applications. In a large cluster, customers would likely dedicate two servers just for the policy engine.
The extra horsepower is needed to support all of the application shifting functions Veritas has planned for Next Generation Cluster. Take, for example, a four server cluster running a database, application server and Web server. Veritas' new clustering server would allow the user to rank these applications, giving say the database top priority and the Web server the lowest priority. If a failure occurred on the database server, the Veritas software would look over the cluster to see if enough excess capacity exists to move the database onto another box. This shift would take into account the spare box's processing speed and network configuration to make sure it's up to the task of running a high-priority database. (Veritas is using two new tools called the Advanced Workload Manager and Group Transition Queue to make this magic happen.)
The Veritas software would also check that the user has permitted the database to run, for example, on the same box as the Web server. If these apps are not certified to run together, then no deal. If no excess hardware is available, the Next Generation Cluster would make a call to Veritas' OpForce provisioning server to go ahead and set up a new hardware box for the database. While this seems like a time-consuming process, the OpForce code has high-end data snapshot tools that can cut the transfer of a database down to a few seconds, Senicka said. At that point, the Ejasent software steps in to track performance data and make sure the database is running at speed.
All in all, a most impressive package. The biggest benefit of all the technology is the potential cost-savings. In traditional clusters, users tend to have at least one box sitting idle just to handle possible failures. In the new Veritas model, all systems would be active with just bits of excess capacity available on some of the servers. Again, if a new server is needed for a serious failure, OpForce will go ahead and provision one. All told, users could see server utilization in a cluster go up from around 15- to 20 percent to around 60 percent or higher.
The Veritas product staff admit that they might not be able to pull all of this off by mid-2005, but, they are, of course, optimistic. A prototype of the software will be ready by the end of this year. Whatever doesn't arrive in 2005 will likely be in a second release of the software that will come about 9 to 12 months later. In the second release, Veritas plans to make it possible to run all of these tools across a cluster with mixed OS servers. Solaris, Linux and Windows are the main priorities at the moment.
In addition, Veritas hopes to be able to take advantage of higher-end OS tools with Version 2. Solaris 10, for example, makes it possible to run a limitless number of virtual machines on a single processor. This means you can basically create clusters on a single server. Likewise, VMware's partitioning technology does similar things for Windows and Linux boxes. Veritas want all of its failover/provisioning goodies to work with this partitioning technology.
These are all lofty goals to be sure, but Veritas appears to have the technology to make it happen. We'll bring you more detail as it arrives. ®