Technology Guide: Improving Server Performance

Related Links

whatis.com

Contents

Caching

Load Balancing and Mirroring

SSL Accelerators

Caching

Caches store copied files and are typically stored on a proxy server located between enterprise users and the Internet.  Caching is implemented by most proxy server products used in enterprises. 

Web caching works on a simple principle: if many users are accessing the same remote web pages, storing a copy of these pages locally (near the user) reduces the consumption of internet bandwidth and speeds the delivery of the information to each subsequent user. These cached pages generally are stored on a proxy server or dedicated caching server located between an enterprise or ISP network and the internet.

Caching is of two types. On demand caching adds documents or objects to the cache only when they are first requested by a user.  On command caching, used by ISPs and corporate intranet proxy servers, preloads the cache with frequently requested pages during periods of low bandwidth use (such as at night).

Load Balancing and Mirroring

A method of distributing (often dynamically) processing and communications across a group of servers or other network resources so that all servers are used to approximately the same extent and no single server is overloaded.  When one server becomes more heavily loaded than the others, the load balancer forwards requests automatically to the least busy server. Requests also can be routed by type of content requested.

Load balancer also can identify specific traffic according to its URL, cookie, or client’s source address.  Load balancing can be handled by a dedicated server, a separate black box, an advanced network switch, or software such as StoneSoft’s StoneBeat.  It can be combined with mirroring.  If the load balancing software operates between servers connected to the Internet or to another wide area network (WAN) at different points, telecommunications costs can be reduced and performance improved by routing requests to the server closest to the user requesting the information. A load-balanced network distributes incoming traffic among replicated servers by means of algorithms such as these:

·        Round robin – all servers are treated equally. Connections are assigned sequentially among the servers in a logical community.

·        Least connections -  the server with the least number of connections gets the next connection.

·        Weighted round robin distribution – individual weight or priority is given to a specific server on the basis of its requirements and capabilities.

·        Maintenance rerouting – if one server is unavailable, traffic is rerouted to another server supporting live fail-over.  Maintenance rerouting creates a fault-tolerant service that rapidly approach 99.99% reliability.

A mirror site duplicates the content either at the same physical location or elsewhere.  Mirroring is done for performance, backup, high availability, and disaster recovery reasons.  A mirror site at the same physical location can share the workload with the original site.  A mirror site at a distant location can both share the workload and provide faster access to users who are far from the primary site.

Vendors have focused on improving their products’ ability to route information intelligently. A number of relatively small vendors are in this market; among them are Coyote Point Systems, F5 Labs, Foundry Networks, HydraWeb Technologies, Network Appliance, Radware, and Resonate.

SSL Accelerators

An SSL encryption accelerator handles confidential information such as credit card numbers.  An SSL-encrypted page can take 2.5 times as long to download as a plaintext page, so using the accelerator relieves the burden on the main Web server.  The slower rate implies that the number of connections per second the server can support is much lower that is the case with plaintext.  An SSL accelerator enables download rates closer to those of plaintext pages.

Intel introduced an SSL accelerator card that improved the performance of a secure web server by a factor of 47.