Research Interests

Here are some of the areas and projects I was involved with. You can read through or jump directly to a topic:

Web-scale computing

I've worked at various Web companies (Microsoft, Facebook) studying how systems scale to hundreds of thousdands of nodes. This includes all aspects from the lowest level hardware (processors and networks), through data centers, and up to the middleware and application layers.
Related publications:

Parallel Job Scheduling

Dr. Dror Feitelson from  the CS dept. at Hebrew University brought me into the field of supercomputing, with his vast knowledge and meticulous research methods. In our work for my MSc thesis, we developed a new method of coscheduling for supercomputers to deal with load imbalances. This work has been extended to a PhD. In this work, we've shown ways to enhace large-scale systems by improving application performance, improve fault-tolerance, reduce system load, and improve resource utilization.
Related publications:

Scalable system software and STORM

One of the main research questions our team is trying to address is how to develop scalable, high-performance system software. As part of this effort, we developed and advanced resource management tool, called STORM (Scalable TOol for Resource Management). This environment was measured to have unprecedented performance in typical resource-management tasks such as job launching in large clusters. STORM is also an excellent platform for studying, implementing and evaluating various job scheduling algorithms, and many of these are already incorporated in STORM.
STORM and system software related publications:

Network Protocols and interconnects

We have investigated and proposed several network protocols for advanced networks that offer multiple rails, that is, a redundancy of networks (interfaces and switches). Multiple rails allow for increased network performance, but hard to exploit efficiently with current bus technology. The papers in this list try to address this with by static or dynamic allocation of rails to messages: