/ Tetsuhiko Ikegami / Vice President
/ Hidesada Kanda / Professor
/ Leo Nagamatsu / Associate Professor
/Akira Fujitsu/Assistant Professor
/ Tongjun Huang / Research Associate
ISTC consists of the Information Processing Center and the University Library. ISTC's main purpose is to provide information services through computer facilities and library services.
Refereed Journal Papers
In the University of Aizu, Information Sysytems and Technology Center (ISTC) has provided many UNIX workstations for student excersises. Students who are beginner users, often logged out a host with unintentionally left processes, and some of these processes may waste CPU resources in many cases. We developed a tool for periodical search for such processes with purpose of basic information for manage and take action. In our university, user files are placed in the central file server and ``on demand'' mounted when a user starts to use a client host. With this type of file mount, it takes a few seconds for ``start up a user process'' even by remote invocation. By this tool with a pipelined parallel processing, to collect process information of each host, we achieved the average processing time for a host of less than one second. We also report the property of found abnormal processes.
Multi-Rendezvous is a powerful communication mechanism that allows a group of processes to execute an event in a synchronous way. Besides the multi-synchronou s property, Multi-Rendezvous has an exclusive property which requires no more th an one group to execute their events simultaneously, if they share a common proc ess. It is not a trivial task to implement Multi-Rendezvous in a network system. The implementation has to maintain the properties mentioned above, be fair and make progress. Some highly abstract specification languages have employed Mult i-Rendezvous to enhance the specification power of the languages, e.g. LOTOS, a specification language for communication protocols and distributed systems. $Coterie$ is a communication structure which has been used in solutions for many distributed problems such as mutual exclusion, replica control, and distributed consensus. Our previous work has shown that $coterie$ can also be used for solving implementation problem of Multi-Rendezvous, with $O(N^{1.5})$ message passes. In this paper, we propose a communication structure called {\it l}-chain-coterie, as a generalization of $coterie$, and a more efficient distributed algorithm for implementing Multi-Rendezvous based on the structure with $O(N^{1+1/l})$ message passes. Our algorithm is fully distributed and does not use manager processes and auxiliary resources.
To establish an effective management system for computer cluster is important, from both educational and research viewpoint. In classroom workstations of our university, some hosts are heavily loaded when no one has logged in. This load is mainly caused by aimlessly running process. In educational situation, owner of these processes must be warned. In research view, tracing load behavior of each host is useful for distributed computing. The total number of classroom workstations is quite large, with many educational computer classrooms of our university. We searched aimlessly running processes, and reported the result of searching experiments for those aimlessly running processes.
Refereed Proceeding Papers
Nagamatsu L. and Suwa D., Searching Aimlessly Running Processes on Classroom Workstations. Proceedings of the 1999 International Conference on Parallel Processing (ICPP) Workshops on Industrial Applications on Network Computing (INDAP), pp.418--423, IEEE Computer Society, Sep.1999.
To establish an effective management system for computer cluster is important, from both educational and research viewpoint. In classroom workstations of our university, some hosts are heavily loaded when no one has logged in. This load is mainly caused by aimlessly running process. In educational situation, owner of these processes must be warned. In research view, tracing load behavior of each host is useful for distributed computing. The total number of classroom workstations is quite large, with many educational computer classrooms of our university. We searched aimlessly running processes, and reported the result of searching experiments for those aimlessly running processes.
Unrefereed Papers
Dialup IP connection facility is to provide users at home an access mean to the campus network. We analyzed usage situation of ports especially in peak time at midnight. By logging transfered data quantity with periodical SNMP check at every 15 minutes, we could estimate various user activity types such as download dominated or interactive sessions. On the distribution of connected duration, pattern of very close to the exponential distribution is found. Based on these record, with queueing theory, the number of ports needed to satisfy specific ratio of connection requests. This result is helpful for future facility design.
Academic Activities
Others
Classroom Workstations in our University are used by students f or educational activities and research. We monitored processes on the classroom workstations, and found some hosts are heavily loaded while no one is logged in. Careless users tend to pay no attentions on presence of unnecessary processes. When the processes are left running long time, in educational situation , where many students share workstations, owner of these processes must be warned. From research view, tracing load behavior of each host is useful for distributed computing. We developed a searching program which examine all host in computer exercise rooms, by scanning every hosts sequentially by remote shell from one host, and listing processes on each host. We regard a process without logged-in owner as aimlessly running. We searched for them during three 4 weeks periods, and properties of detected process are reported. To develop effective user warning system, we studied to select most serious in performance processes only. Based on load value of a host and process status, we can focus into 0.5% of the total number of aimlessly running processes. This system is useful for the system administrator to keep the computers in good condition.
Distributed Web Server Systems (DWS) provide a way to improve WWW server performance, by distributing requests from clients to several servers which jointly process the requests. Many variations of Distributed Web Server (DWS) are known, such as a system based on Domain Name Server (DNS) approach. The features of DWSs are related to user transparency, scalability, portability, possibility for geographically distribution, and stability on request bursted situation. In this paper, we introduce A WWW Request Distribution method by URL rewriting feature of the Apache Server. This method uses the built-in module of the Apache Server, so as to provide ease of usage. Request Distribution method does not require any change of the Apache source, and easy to implement with portability. This method can work with DNS distributed system. To maintain load balancing, combination of DWS and Request Distribution method is useful. In this research features of distribution methods are compared. For proposed combined method of DNS and URL rewriting, performances are evaluated, improvements on server load reduction is shown. With this method more flexible load distribution is achieved.