Performance evaluation of Web Cache

Next we will consider an example of discrete event simulation implemented in CSIM. The example discusses the effect of Web caching on network planning in sense of determining the bandwidth of the access link interconnecting the ISP’s subnet with the internet. This is studied by means of simulations [1]. The latency of a browser retrieving files is studied for given traffic characteristics, number of users, bandwidth of access link and cache hit ratio. We divide our study into two cases. In the first case we assume the arrival process to be Poisson and service times to be generally distributed. In the second case we assume that the arrival process is an ON-OFF process, with sojourn time in ON state being Weibull distributed and in OFF state being Pareto distributed. This model represents typical network infrastructure interconnecting a subnet with the internet. Browsers access the subnet via Modem, ISDN BRI, Ethernet or Token Ring. When a WWW browser clicks on a hyperlink, several URL requests may be sent from the browser to the web proxy which may be just one cache or a collection of caches. If the proxy has a copy of the requested file and is consistent with the original copy on the remote server, the proxy sends the copy to the browser. This is called a hit. If the proxy does not have a copy of the file that the browser is looking for, the proxy retrieves an original copy from the remote server, sends it to the browser and keeps a copy in the cache. This is called a miss. In this example we are interested in file delivering latency or mean response time, which is defined as the time interval from the browser clicking an object to the requested object being displayed on the monitor. Excluding some trivial terms such as the delay of monitor display and cache retrieval and the HTTP interaction within the subnet, the mean response time consists of the following terms: