Electrical Engineering-Systems Dept.

סמינר מחלקתי

You are invited to attend a lecture by

Dr. Eli  Brosh

(Director of Engineering, Vidyo, Inc.)

on the subject:

Scalable Virtual Machine Streaming and Distributed Replication Systems 

Clouds commonly store Virtual Machine (VM) images on networked storage. This poses a serious potential scalability bottleneck as launching a single fresh VM instance requires, at minimum, several hundred MB of network reads. As this bottleneck occurs mostly severely during read-intensive launching of new VMs, we focus on scalably minimizing time to boot a VM and load its critical applications.

While effective scalable P2P streaming techniques for video-on-demand (VOD) scenarios where blocks arrive in-order and at constant rate, are available, no techniques address scalable large-executable streaming. VM execution is non-deterministic, divergent, variable rate, and cannot miss blocks.

To address these issues, we develop VMTorrent, an effective solution for scalable VM streaming that is based on making better use of existing capacity, instead of throwing more hardware at it. It introduces a novel combination of block prioritization, profile-based execution prefetch, on-demand fetch, and decoupling VM image presentation from underlying data-stream. VMTorrent 

Supported by analytic modeling, we present comprehensive experimental evaluation of VMTorrent on real systems at scale. We find that VMTorrent supports comparable execution time to that achieved using local disk. VMTorrent maintains this performance while scaling to 100 instances, providing up to 11x speedup over current state-of-the-art.

In the second part of the talk, we consider the problem of how to place and efficiently utilize resources in network environments. The setting consists of a

regionally organized system which must satisfy regionally varying demands for various resources. The operator aims at placing resources in the regions as to minimize the cost of providing the demands. Example of systems falling under this paradigm are content service platforms and cloud computing services.

The main challenge posed by this paradigm is the need to deal with an arbitrary multi-dimensional (high-dimensionality) stochastic demand. We show that, despite this complexity, one can optimize the system operation while accounting for the full demand distribution. We provide simple-yet-efficient algorithms for conducting this optimization. The importance of the model is demonstrated by showing that an alternative analysis which is based on the demand means only, may, in certain cases, achieve performance that is drastically worse than the optimal one.