REVIEW, Lessons from Giant-Scale Services

The paper talks about some topics related to giant-servers including load
management, high availability, graceful degradation, and updating. In
relation to each topic, the author introduces some techniques, and
especially in area of high availability, he describes some metrics. He
compares different techniques in replication and partitioning. Also,
describes different approaches to upgrading giant-servers.

The key strength of this paper is that the author is one of the founders
of Inktomi company. Therefore, he has many experiences related to these
kind of servers and practical problems in working with them. In Inktomi,
he had a great chance to test many of the mentioned ideas in practice.
Another good point in this article is introducing metrics for measuring
high availability and service capacity of servers.

Although described many techniques, but the rule of application type and
types of services, being offered by these servers, is understated. For
example, different techniques in load management and replication and
partitioning can be used for stateless and stateful services.
In addition, there are many other techniques in each of those areas, which
could be described by the author, but they have not been mentioned. For
example, the idea of having servers with same IP address for load sharing.
Also, he could emphasis more on multi-tier technique and discuss it more
as it is widely used.

About the smart clients, I think, benefits of using smart clients are
highly dependent on type of the application. However nowadays
smart-clients are not preferred in many types of internet services.
Received on Mon Sep 19 2005 - 10:54:47 EDT