Kais Belgaied's Weblog at Sun

Wednesday August 29, 2007

Latency matters I recently went across a bridge in Austin, Tx, and there were two speed limits: a minimum (45 Mph) and a maximum (65 Mph). This made sense, since the bridge has a limited capacity, and, allowing slow vehicles will result in crowding it, and creating a bottle neck. The Internet is often compared to the highway system, and this example actually applies here too. Web transation involve mutiple tiers of servers in the data center, involving a front end tier (or a web tier) and a set of other servers collaborating in handling the transaction. These may include access to an authenttcation service, a database service, and a number of other applications in the data center. Some of these maintain a state for the whole transaction, or for phases thereof. Resources are occupied by these state contexts while the transaction is executed. The higher the latency between various collaborating nodes, the longer the resources are occupied, and consequently the lesser transactions the overall distributed system can accomodate. Latency is usually associated with the response time of a single transaction, whi results in a quicker execution of a trade order for example, or a shorter waiting time for a use. It turns out that latency actually plays also a role in the scalability of distributed systems capacity, and indirectly affects resources provisioning. Now a disclaimer about the bridge example: I am in no way advocating exceeding the driving speed limits with the pretext that it helps reduce traffic congestion. Posted by kais ( Aug 29 2007, 05:23:02 PM PDT ) Permalink Comments [1]