Regularly expected reference-time as a metric of web cache replacement policy

• Main Author: Sediyono, Agung
• Format: Conference or Workshop Item
• Language: English
• Published: 2011
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://repo.uum.edu.my/13660/1/287.pdf; http://repo.uum.edu.my/13660/; http://www.icoci.cms.net.my
• Institution: Universiti Utara Malaysia

The growth of Internet access was increasing significantly.In facts, more than one user access the same object so there is an opportunity to reduce this redundancy by placing an intermediate storage called cache.By this approach, the bandwidth consumption and response time of system in term of user perception can be improved. When the size of web cache is limited, it needs to manage the objects in web cache so that the hit ratio and byte hit ratio are maximized. Based on previous research the performance of cache replacement is dependent on the user/program access behavior.Therefore, the success of IRT implementation in memory cache replacement is not guaranteed a same result for web cache environment.Researcher has explored the regularity of user access and used this characteristic to be included in a metric of web cache replacement. Other researcher uses the regularity to predict the next occurrences and combine with past frequency occurrences.In predicting process, they use statistic or data mining approach.However, it takes time in computing prediction process. Therefore, this paper proposes a simple approach in predicting the next object reference.This approach is based on assumption that the object could be accessed by user regularly such DA-IRT that be used to calculate the time of next object reference called the regularly expected reference time (RERT).The object with longer RERT will be evicted sooner from the web cache.Based on experiment result, the performance of RERT is dependent on user access behavior and opposite of DA-IRT policy.