Análise Estatística do Trafego de E-Mail para o Projeto de um Cluster de Alto Desempenho
Resumo
This paper shows an e-mail statistical analysis and the behavior of its users based on real servers utilization. The work is related to an specification project, under development in lnstitute for Technological Research (IPT-Brasil), for a high scalability and availability e-mail service ( above 100.000 users). The statistical results are a important step to the system performance evaluation, allowing an implementation of realistic workload simulations. The data collected are related in three categories: the amount of messages per user, the time interval between messages and the size of each message. The time intervals and the size of messages are calculate in accordance to a model thats use probability density functions fitting the real data, scaled to desired quantity of users. To carry through the simulations, Java applications will be implemented to generate messages with the results that best fits the realistic workload. The work intends to simulare the requisitions generated by the users, to measure the system capacity according the servers quantity and performance, to characterize servers utilization pattems and users behavior in arder to determine the restrictive performance factors.
Referências
MENASCÉ, D.A. "Loading Testing of Web sites", IEEE Internet computing, IEEE, Jul./Ago. 2002.
BERLOTTI, L; CALZAROSSA, M.C. "Models of Mail Server Workloads", Performance Evaluation 46, Elsevier, pp 65-76, 2001
BERLOTTI, L; CALZAROSSA M.C. "Workload Characterization of Mail Servers", Proceedings of SPECTS2000, Vancouver, Canadá, Jul. 2000.
ZHANG, Wensong,, "IPVS Home Page", Internet, Wensong, " http://www.linuxvirtualserver.org ", Acessado em 9 jan. 2005.
ROBERTSON, A, "Highly-affordable High Availability", Linux Magazine, Nov. 2003.
AMORIM, A.C. O., "Ajuste de Funções com Algoritmos Genéticos", disponível em [link], Acessado em 1 set. 2005.
SAITO, Yasushi; BERSHAD, Brian; LEVI, Henry. Manageability, availability and performance in Porcupine: a highly scalable, cluster based mail server. University of Washington. Dez. 1999.
BEHREN, J.R.V., CZERWINSKI, S., Joseph, A.D,. BREWER, E.A., KUBIATOWICZ, J., "NinjaMail: The design of a High-Performance clustered, Distributed E-mail System", IEEE, 2000.
MITCHELL, M., "An lntroduction to Genetic Algorithms", Cambridge : MIT, 1996
MILONE, G, "Estatística Geral e aplicada", Thomson, São Paulo, 2004.
MENASCÉ, D. A., ALMEIDA, "Planejamento de Capacidade para Serviços na Web", Campus, Rio de Janeiro. 2002.
