Data Center Benefits

Perfect Search technology provides the ability for a company to dramatically reduce the number of servers required to handle its search volume.  To better understand this, it is useful to understand the distributed architecture of low-end servers that many companies use for search.

Server Farm Clustering

Your browser may not support display of this image.A computer cluster is a group of linked computers, working together closely so that in many respects they form a single computer. The components of a cluster are commonly, but not always, connected to each other through fast local area networks. Clusters are usually deployed to improve performance and/or availability over that provided by a single computer, while typically being much more cost-effective than single computers of comparable speed or availability.1

Generally, server farms are comprised of lower end machines.  A typical server in a server farm might have 4 gigabytes of RAM in memory and have a 7200 rpm drive.  Some companies, like Microsoft, will try to gain query speed by caching the entire index, while others will only try to cache their most popular queries.2

Within a server farm, there can be a number of clusters that perform different functions.  The servers we focus on are the query servers.  These are the servers that are dedicated to finding the data in the index and returning the results set.

A cluster group of query servers can be wide, based primarily upon the size of the index.  If the index is very large, the cluster group will be wider.  In order to get the most speed as possible, companies will subdivide the index into segments, so that portion of the index will be able to sit in cache as much as possible, since retrieving information from cache is much faster than retrieving information from disk.

For example, if a company has an index of 100 GB and the query servers have 4 GB of RAM, and they want to have the entire index cached in memory, it would need to segment the index into at least 25 segments (100 / 4 = 25) and store those index segments on 25 separate servers.  This would be a cluster grouping with 25 subgroups wide.

Each subgroup can also be deep.  This is primarily a function of the query volume.  Assume that a typical query server can handle 20 queries per second.  If a company has a query volume of 1000 queries per second, then each cluster subgroup would need to have 50 servers to handle that query volume (1000 / 20 = 50).

So if a company had an index size of 100 and a query volume of 1000 queries per second, given the typical server configuration, they would need to have a query server cluster of 25 cluster subgroups with 50 servers in each subgroup.  The total number of query servers needed would be 1250.

It is estimated that companies like Google have between 300,000 and 500,000 servers in their server farms as they handle the enormous size of the indexing of the internet and handle the tremendous query volume.

Perfect Search Server Reduction

Because Perfect Search can get query speeds orders of magnitude faster than traditional search engines, it means that more queries can be handled per server with the PS Search Engine. By having an increased query throughput per server, fewer servers are needed to handle the same volume of queries as traditional search engines. In our tests, we have seen that we can achieve sufficient increases in query speeds effectively to replace between 10 and 1,000 servers by a single server using Perfect Search. That means that we could handle the query volume on 1 server that it would take 10 servers to handle with a traditional search engine.

If we were to take the example above in which each server could handle 20 queries per second, it would take 50 servers in a cluster subgroup to handle a query volume of 1000 queries per second. If Perfect Search could handle 200 queries per second per server, that would mean that we could replace the 50 servers with just 5 servers to handle the total query volume.

More on Server Farm