CSE523 Projects Available with Professor Stark

• A Network Memory Server

A Network Memory Server

The past history of this project is that we built a Network Memory Server on a cluster of Alpha and Intel machines. What was lacking from this project were motivated contributors who wanted to see their applications run on the platform. Consequently, the hardware platform for this system has been mothballed for the time being. However, I still think there are some interesting directions to pursue (see below).

Current Goals (Spring 2007)

My current interest in this project is to follow through more completely on some ideas I had for predicting paging requests. A student, Ahmed Burukan, made significant progress on this in 2004/2005 (see below), but the project did not reach the point of an interesting research publication. I think it would be interesting to try these ideas out either in the context of normal paging on a single workstation, or else in the context of the memory management subsystem of a language run-time system such as that of Standard ML of New Jersey. In the latter case, it might be possible to do this without operating system kernel modifications, but more likely some minimal modifications would need to be made to add some "hooks" to the page-fault handler. Any suitable open-source operating system (such as FreeBSD or Linux) could be used for this.

If you are interested in learning more about this project, please see the SAMSON Project Home Page. The up-to-the-minute source code can be found in the SAMSON source code CVS repository. E-mail Professor Stark if you want to discuss possibilities.

Project History

• Fall 2004/Spring 2005: Ahmed Burukan investigated an idea for predicting future page accesses from past history, and implemented a simulator in Java for analyzing the performance of this prediction method on actual traces from the visible human application.

• The basic system is functioning and can run a "Visible Human" volume-rendering application. It is possible to extract detailed information about applications running on the system.

• Early Summer 2003: Professor Stark implemented the event logging mechanism that was part of the original design. It is now possible to collect detailed traces of all aspects of the operation of the system. See here for a detailed look at the system performance on one of our test exercisers. See here for a report on the "Visible Human" application.

• Spring 2003: Project stalled for lack of interested students.

• Fall 2003: Harikesavan Krishnan joined the project, then left it. Prof. Stark did some further debugging work on the system and merged the last code that Eric Nuzzi had worked on in 2001, with the result that the system now functions in its target configuration with the Alpha systems as clients and the Compaq cluster as servers.

• Spring 2002: Project is currently stalled because I don't have students who want to work on it. It's a good opportunity for someone who wants a challenging project they can make their own.

• Summer/Fall 2001: Not too much happened on the project. Eric Nuzzi told me he has decided to leave school for now. I received code for a potential test application from Susan Frank, who is working under Prof. Kaufman. This code is voxel-based code for rendering 3D images of the "visible human" database. The code looks to me like it's got good potential as a test application, but it needs somebody to work on it and on the system. I want to get a paper out of this project, but I need performance numbers. If you can help me get these numbers, you can be an author, too.

• Spring 2001: Eric Nuzzi implemented some of the reliability/replication code and ran some more rigorous system exercise tests.

• January 2001: A rack-mounted cluster of 7 Compaq ProLiant servers was assembled by Professor Stark and Eric Nuzzi. Each server is an 933MHz Pentium III system with 3GB of RAM, a 9GB SCSI drive, a 100Mb/sec fast ethernet interface and a Myrinet interface.

• Fall 2000: Further development of the prototype by Professor Stark and student Eric Nuzzi. The code was moved to a more recent 2.2 version of the Linux kernel and the old 2.0.22 version was deprecated. The system was updated to work with arbitrary combinations of 64-bit Alphas and 32-bit PCs as clients and servers.

• End of July, 2000: The system had been debugged to the point that it was beginning to function reliably on tests using realistic amounts of network memory. A significant "push" is still needed to remove remaining serious bugs and extend the implementation so that actual applications can be run.

• Fall 1999 - Summer 2000: System prototype coded and debugged by Professor Stark. Supports basic functionality, which basically consists of ability to memory-map and access network memory. Scaffolding is provided for most of the features discussed in the design document, but only the essentials are currently implemented.

• Fall 1999: Four Alpha 21164 workstations, each with 1GB RAM, together a Myrinet switch and several PCI interfaces, were purchased for system development. Development commenced using these systems and on previously existing Pentium 200MHz systems, running under Red Hat Linux, kernel versions 2.0.22 and 2.2.12.

• Spring 1999: System design done by Professor Stark and several students in course CSE 679. The result was the following design documentation. This document describes the general structure and function of the system. It is now somewhat old, but the only newer information is the source code itself.