Sizing up servers
- Article 8 of 77
- Information Age, May 2001
Will IT managers ever find a way to escape the constant round of server upgrades?
Page 1 | Page 2 | All 2 Pages
Intel's Priestley adds that while rack-mounted servers may bring additional expense at the outset, they can actually be cost-effective in the long run, particularly when problems occur. “It might take four hours to have someone come in and fix a broken server. The question is, is four hours downtime acceptable to a business? If it's not, then simply being able to replace a server is a good thing.”
Reliable evidence?
As yet, there is no evidence to suggest that rack-mounted servers are less reliable than their more conventional cousins, and, as they are made from industry-standard parts, they can work out cheaper than a tower chassis, with only the initial outlay for the rack needed. Indeed, many of the reliability issues are more relevant to tower-chassis servers because a single server presents a single point of failure: it is very noticeable if a single server stops working when it is the only one providing a service. But if there are 40, it does not matter so much if one falls over.
So at what point should a company switch from tower-chassis servers to rack-mounted servers? According to Ian Meakin, the UK enterprise server marketing manager for Sun, there is no clear-cut point at which to cross over. “It depends on the organisation's applications and its market. I don't like to think of it as an 'either/or' situation. We usually send a team in to discuss the company's architecture and what its requirements are, rather than say it must have one or the other.” Sun's top range Fire server is rack-mounted, using a high-bandwidth (9.6GB/s) backplane to achieve the same speeds as an internal bus for data communication between servers.
Meakin's argument also goes for companies that are looking to buy a mid-frame server that could rival the capabilities of several smaller servers. “A large Oracle database requires 40GB of memory so a smaller server isn't going to be able to cope with that. You need a massively, vertically -calable server, rather than a series of horizontally scalable servers as you would for web provision.”
While clustering small servers together is an option for load-balancing and failover provision, it cannot support the requirements a large database application places on a single server, Meakin argues. So no matter how good the argument is for smaller, rack-mounted servers, there are some applications that absolutely require a single, highly powerful server.
A company needs to look at what applications it intends to deploy on its servers before deciding how to handle its server needs. It may eventually settle on a hybrid approach - highly scalable servers with considerable degrees of redundancy and failover for database applications and other programs that need to have single, powerful servers; and racks of thin servers that are far less powerful individually, but together are capable of responding to far larger numbers of transactions, with multiple failover points.
Most server vendors are looking at the dense, rack-mounted server carefully, as the potential it offers to medium and large enterprises is clear. But the single, highly-powerful server is still invaluable for the enterprise that needs raw power in a single box.
Page 1 | Page 2 | All 2 Pages