Subject: Long jobs history #5 Date: Thu, 08 May 1997 17:46:35 EDT From: "Naomi B. Schmidt" ------- Forwarded Message Received: from SOUTH-STATION-ANNEX.MIT.EDU by po6.MIT.EDU (5.61/4.7) id AA03825; Tue, 8 Aug 95 12:03:46 EDT Received: from MOZART.MIT.EDU by MIT.EDU with SMTP id AA10266; Tue, 8 Aug 95 12:03:34 EDT Received: by mozart.MIT.EDU (5.57/4.7) id AA26880; Tue, 8 Aug 95 12:03:34 -0400 Message-Id: <9508081603.AA26880@mozart.MIT.EDU> To: patriot2@MIT.EDU Cc: gjackson@MIT.EDU, cec@MIT.EDU, rar@MIT.EDU Subject: Notes from the 8/7/95 Long Jobs meeting Date: Tue, 08 Aug 1995 12:03:34 EDT From: "Naomi B. Schmidt" Present were Matt, Carla, Tom, Steve, Miki, and Naomi Miki reported on her digging into the historical lore of Patriot1. - We tried to make the VAX9000 into as much of an Athena workstation as possible. - The way it worked was that the command 'patriot' on any Athena workstation logged you on to the machine. Once on the VAX9000, you could use the command 'submit' to submit your job. - There were many bugs in submit, which we fixed. - The hacked kernel caused Patriot to frequently crash. - People had lots of complaints about their 'nice' levels. - Depending about the load on the machine at any particular time, a compute intensive job might take different amounts of time each time it was run. - People were running all sorts of jobs on it, from IRC to mail, to whatever John Carr was doing. There was no tracking or record keeping. - We gave people 10 hour tickets by default. Our next steps are as follows (to be done by August 28, at which time we will meet again): - We will recreate patriot on two low-end SPARCs (Classics or SPARC5's) (Matt will take over from Miki with Yoav's help) - We will port dialup to Solaris so as to plug security holes. This includes a sane version of sendmail so that output can be sent to users (Matt) - Concurrently, we will do some benchmarking on a SPARC20 with two processors to find out what the throughput looks like (Steve Ellis) - We will contact Dennis Aylward to get a quote on a SPARC20 with 4 processors, equipped as in previous notes (Naomi) - We will identify a pilot audience for the Fall and solidify our policies on rules of use - who is allowed to use it, how people register to use it, etc (Naomi) Other discussion of details: - We will give 23 hour tickets by default. - We will give Matlab the IP address of the machine to allow it to continue running without renewal of tickets. - We will keep logs of processes that are run on the machine and how many CPU minutes/hours each takes. - The scheduler will 'renice' jobs to some extent to do some queue management. Other ideas: - Rather than allowing users to log into patriot, we might write a script to do rkinit, so that they just submit jobs to it - that way they can't just log in to read mail when the dialups are busy. - The service would be available to two classes of users: - people in classes that register to use the machine - instructors will register with us and will maintain the lists of individual usernames - people who need it to do Thesis or UROP work. They will fill out an electronic form stating what they need it for and we will add them to the password file. The form will state explicitly what the machine may and may not be used for - violation of this will allow us to remove an individual's access. [Idea - modify the 'register' program so that they can do this and be added to the password file automatically] Criteria for Success: - -------------------- 1. We will gather stronger data on the magnitude and nature of the use of such a facility. 2. We will be able to decide whether a service such as this belongs in the environment, possibly with a modified technology such as a commercial batch software package. 3. We will know whether there is a reasonable way that we can satisfy the magnitude of the demand. 4. We will have a better definition of the support requirements for such a system. 5. People will leave fewer workstations logged in and tied up with no one present. 6. The faculty say that we are addressing their instructional needs. ------- End of Forwarded Message