From theory to practice
- Article 1 of 1
- The promise of server consolidation, February 2002
How should a company that wants to consolidate its servers go about the process?
Page 1 | Page 2 | All 2 Pages
Sun, in turn, offers similar capabilities on its Sunfire servers, having fault-tolerant clock boards, hot-swappable clocks and memory boards. Unisys's ES7000, meanwhile, can “lose memory, lose processor cores, lose complete IO modules,” says Ian Benn. It even has two mains cables and two systems clocks in case one of either develops a fault.
All the servers offer considerable amounts of data throughput since the large number of processors in servers of this size requires a large amount of data to be supplied at any one moment to maintain full utilisation. Lewis boasts that the Sunfire 15k can transfer 43GB per second of data to the processors. Tikiri Wandarugala, senior server consultant of IBM Europe, counters that the Enterprise X-Architecture has its data interconnection built into the memory controller rather than the bus and so is capable of scaling to higher levels.
Indeed, IBM claims the unique selling point of its X-Architecture is that customers can increase the capabilities of their server at any time. For example, they can add a second server to the first and create a box with double the processors instead of just two servers. They can even switch tasks, converting an 8-way into two 4-ways and back again with a reboot of the operating system.
Partitioning is another critical technique for consolidating multiple low-end servers into one or two high-end machines. Using partitioning, a single server can be made to look like multiple servers with different operating systems and resources. “Partitioning is absolutely vital,” argues Lewis. Applications can be placed on different virtual servers according to their requirements and mission critical apps can be distributed so that if one operating system collapses, the server and the other operating systems continue unaffected. Lewis claims Sun is the only company to offer dynamic partitioning, enabling servers to reallocate resources depending on load or according to a schedule. An application that runs its server at 80% utilisation can automatically be allocated more processors to match demand. “You can set the machine to be completely self-reliant,” he says.
Nevertheless, Lewis says that no one would necessarily want to consolidate onto just one machine. “I don't think one server can do everything anyway,” he confesses. Alan Priestley, strategic marketing manager for Intel's enterprise marketing group, points out there may be problems of scalability for applications, for instance. “An application might run great on a 4-way platform, but what happens if I put it onto a 32-way platform? What will my back-up strategy and recovery strategy be when I have to back everything up in one go from one server? Do I have the network bandwidth to deliver data out to users from one site if I used to have servers closer to the user?”
Priestley advises clustering groups of larger servers to consolidate at least some of the small servers, but to avoid the issues he highlights. “Failover clusters are pretty common. The Oracle 9i database system supports failover clusters. And the clusters can offer greater availability than a single server – six 9s possibly, whereas the individual servers may offer only five 9s,” referring to systems that support 99.999% up-time. But Unisys' Ian Benn points out that some applications cannot be partitioned or broken into chunks so need to be on a single server.
For companies intending to consolidate servers, Priestley advocates involving a systems integrator or hardware supplier very closely. But Benn suggests that the project should be treated as a small mainframe project rather than a large Windows project if the company wants to succeed. “Most 'Windows projects' fail. Mainframe thinking doesn't go wrong.”
Page 1 | Page 2 | All 2 Pages