# Paper 91 - Enhancing the Web's Infrastructure - From Caching to Replication
1) Problems to be addressed/ Objectives
This paper proposes a design strategy that seeks to address two of the most challenging and important issues facing the users of the World Wide Web
today - i) Improving on the latency of data retrieval, by decreasing the document fetch times ii) Improving the efficiency of Bandwidth
Utilization The paper aims to improve on these factors by a change in the Network Infrastructure - not the hardware, but the organization. However,
the change in the infrastructure is such that it does not necessitate any modifications to existing systems or software. This helps in improving the
performance statistics, without a major upheaval incorporating the changes, thus keeping the cost of change to a minimum. The design proposed also
aims to keep the following in mind. i) Reduction of the amount of data transferred ii) Improving availability of the document, in other words,
server uptime and performance iii) Transparency of the system - The changes made to the system should not be externally visible
2) New Ideas/ Strengths
The design takes two traditional ideas of improving performance - caching and replication - and combines them in a way that achieves much better
results. Some of the strengths of the design are as follows. i) It takes into account the document size, the pattern of access for a document, the
time required to make a second fetch for the document etc while considering the strategy for caching and replication. This is a very valid point,
because it would be much better to make a cache hit on a single 10 MB file that has a than to make a cache hit on 10 files of 100 KB, on the
assumption that the frequency of access for both instances were to be the same. A combined strategy will result in better performance. Also, the
system is transparent - the design modifications are not visible to the user. ii) The performance statistics are based on the "byte hit rate" and
not merely a cache rate. hit iii) The CgR/WLIS design addresses the issue of "document staleness" in a simple, yet effective manner. It utilizes
times of low traffic to maintain data consistency. iv) The structure of the Replicated Servers and their interconnections is very simple and clean
- and so is the information propagation and "data buildup in WLIS". v) "Balance in Bandwidth utilization" by using the network at times when
traffic is less. vi) "Failure Tolerance" or increase in "Service Availability", because of the replication at various places. vii) "Traffic
Shaping" seems to be another new feature of the design.
3) Weaknesses/ Extensions
Some of the weaknesses of the design are listed below. i) "Document Retrieval Latency" in cases where a copy of the document is not replicated
would be quite high, because an initial search will have to made through the replicated servers in a hierarchical fashion, leading only finally to the
primary server. Therefore, the performance of the system will depend to a great extent on what the "hit rate" is - and how much of a gain it is. The
effectiveness will depend on the tradeoff between the...