Jim's Pages

Software Architecture  

Architecture Trends   

[Architecture Home]

   
  Application Trends

CommodityServers

Grids are "clusters of clusters". Built up as farms of commodity servers they represent an alternative to the large chassis full of SMP (Symmetrical Multi-Processing) blades. Along with the usual high availability benefits of normal clusters, such grids support fine grained scalability tuning by adding or removing a few or many servers as needed.

The motivators all deal with money. Yahoo or Amazon (I'm still looking for a link to help me remember which) did detailed computations of power consumption, heat emission, floor space, deployment, reliability and management and concluded that "more and smaller" servers made the best sense. They have such large farms that they expect to "lose" one or two servers every day to unrecoverable hardware failure. On pure cost/benefit reasoning, it's not worth the effort to pull the dead ones out of the racks. They just add more on the other end.

A LAMPInfrastructure makes it easier and cheaper to manage huge numbers of servers.

 Who's Doing This?

From the IT Manager's Journal: "Google is now rumored to be running more than 100,000 Linux servers, and laying plans to leverage its server infrastructure in ways that extend far beyond search."

From ZDNet: "Amazon said in June [2001] that it was revamping its computer systems and switching to "commodity" computers running Linux. Executives said at the time that they expected technology costs as a portion of net sales would decrease by 20 percent this year.

Architecture Trends   

[Architecture Home]

 
Copyright 2001-2005, Jim Standley