Guangyu Shi

Microwiper: Efficient Memory Propagation in Live Migration of Virtual Machines

Live migration of virtual machines relocates running VM across physical hosts with unnoticeable service downtime. However, propagating changing VM memory at low cost, especially for write-intensive applications or at relatively low network bandwidth, is still uncovered. This paper presents Microwiper, an improvement of memory propagation in live migration. Our idea is twofold. We propose ordered propagation to transfer dirty memory pages according to their rewriting rates. We factor available network bandwidth in sending pages to throttle hot spot; after the accumulated rewriting rate exceeds the estimated bandwidth, next iteration is started immediately. The combination of these novel methods can not only reduce dirtied pages, but also shorten service downtime and total migration time. We implemented Microwiper by retrofitting the pre-copy approach in Xen hypervisor. We conducted detailed experiments to evaluate its efficacy on various workloads. The experimental results show that Microwiper can significantly reduce downtime and transferred pages by more than 50%. Microwiper has good adaptivity, and hence can be applied to other virtualization platforms easily.

SandStone: A DHT Based Carrier Grade Distributed Storage System

Peer-to-peer (P2P) systems have drawn a lot of attention from telecom research communities, due to their advantages such as high scalability and cost-effectiveness. Many P2P inspired architectures have been proposed to implement the traditional telecom functionalities such as rendezvous service and even wireless networking. Although DHT is always adopted as the major element of such architectures, it’s seldom discussed how to make DHT to meet the telecom infra-structure performance requirements, so-called the “Carrier Grade”. In this paper, we discuss the main design aspects and benchmark measurements of SandStone, a DHT based key-value storage system with traffic localization, strong consistency, high availability and scalability, so as to fill the performance gap between telecom infrastructure and general DHT.

A DHT Key-Value Storage System with Carrier Grade Performance

The Peer-to-Peer (P2P) technology being widely adopted in todays both academic research and practical service providing, has many potential advantages and achieves a great success in Information Technology scope. Recently some researchers have proposed that P2P inspired architecture also might be one choice for the telecom network evolution. Most of such works adopted structured P2P (DHT) as the basic solutions, but they seldom discussed how to eliminate the huge gap between the telecom underlay performance requirement and the performance of existed DHT which mainly originated from the Internet applications. This paper presents the design and implementation of SandStone, a DHT based key-value storage system with carrier grade performance, such as good scalability, strong consistency and high reliability, which could be deployed as the cornerstone in such new P2P inspired networking architectures.

Towards a Better Understanding of Locality-Awareness in Peer-to-Peer Systems

Peer-to-Peer(P2P) applications such as BitTorrent and Coolstreaming ignore inter-AS traffic costs at ISPs and generate a large amount of cross ISP traffic. As a result, ISPs always restrict the BT application to control the inter-AS traffic cost. In this paper, we designe a new locality-awareness selection approach: multi-metric neighbor selection algorithm, to enhance local traffic. Based on pwhois query, the algorithm selects the neighbors within the same AS and district as source peers, and uses roulette method based on link bandwidth and latency to choose the final partner peers. Redundancy traffic calculation were implemented to prove the effectiveness of our approach. We evaluate two widely deployed P2P system: BitTorrent and CoolStreaming, under the real dataset from PlannetLab in our evaluations. The results demonstrate our algorithm effects both improving the download time and reducing the inter-AS traffic. Further more, the network congestion can be eased under our algorithm.

Predicting the behavior of large scale P2P systems by parallel discrete event simulation

P2P systems are becoming the dominator of Internet. Such systems are typically composed of thousands to millions of physical computers, which make it difficult to predict their behaviors without a large scale distributed system simulator. This paper is an attempt to predict the behavior of large scale P2P systems by building a novel parallel simulator: AegeanSim, which provides parallel discrete event simulation of such systems on high performance server clusters. We abstract the execution of P2P applications with a specific event model, and parallel the simulation of events in a cluster, thus expanding the simulation scale and boosting the simulation process dramatically. A 1-stage synchronization method is proposed to improve the performance. An event callback interface is designed to implement their application logic easily while keeping the simulator application-independent. We use AegeanSim to predict the behavior of a typical P2P system: BitTorrent. By comparing the simulation behavior of BT with that of related BT studies and verifying its efficiency, scalability and accuracy, we make predictions about the behavior of BT system assuming they are assaulted by different man-made system attacks. Some reasonable results are found: (1) Tracker isolation can hardly work because of its short attacking time window. (2) Limiting the bandwidth of BT may be an efficient way to control it.