lin-mutex - A Practical Distributed Mutual Exclusion...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
Shi-Ding Lin Microsoft Research Asia i-slin@microsoft.com Qiao Lian Tsinghua University lq97@mails.tsinghua.edu.cn Ming Chen Tsinghua University cm01@mails.tsinghua.edu.cn Zheng Zhang Microsoft Research Asia zzhang@microsoft.com Abstract Mutual exclusion is one of the well-studied fundamental primitives in distributed systems. However, the emerging P2P systems bring forward several challenges that can’t be completely solved by previous approaches. In this paper, we propose the Sigma protocol that is implemented inside a dynamic P2P DHT and circumvents those issues. The basic idea is to adopt queuing and cooperation between clients and replicas so as to enforce quorum consensus scheme. We demonstrate that this protocol is scalable with system size, robust to contention, and resilient to network latency variance and fault-tolerant. I. I NTRODUCTION One of the fundamental primitives to implement more generic systems and applications on top of P2P DHTs [4][13] is mutual exclusion. Such primitive is also a rudimentary service to guard arbitrary resources when necessary. For example, a concurrency control mechanism is obviously needed for a mutable distributed file system. For the applications and systems we envision to be built and deployed on those P2P DHTs, one can all but rule out the possibility of enforcing concurrency using stable transaction servers, whether they are external or internal to the system. Therefore, such primitives must be implemented inside P2P DHT. The protocol is thus by definition distributed, it must be simple and efficient, and yet robust enough to be of practical use. Our basic idea is simple: utilizing the fact that nodes in the DHT collectively form a logical space that does not have holes, institute a set of logical replicas upon which a quorum consensus protocol grants access to critical section (CS). From a client’s perspective, these replicas are always online. However, they may suffer from complete memory loss from time to time. Such random reset occurs when the node that acts as a logical replica crashes and gets replaced by one of its logical neighbor in DHT. The open and dynamic nature of P2P environment brings another serious challenge. Many previous approaches [10] assume a close system with fixed and relatively moderate number of nodes, and nodes communicate among themselves to reach consensus. These solutions are inapplicable in our context where the number of clients is unpredictable and can swing to be very large. The protocols need to be designed for a harsh and open environment such as a wide-area P2P. The work in this paper presents a few novel contributions: We start by investigating a straightforward, ALOHA-like strawman protocol and show that, the high variation of network latency between clients and replicas is responsible for the large performance degradation. We believe this insight is valuable for any wide-area consensus protocols. We
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 12/08/2011 for the course CS 525 taught by Professor Gupta during the Spring '08 term at University of Illinois, Urbana Champaign.

Page1 / 6

lin-mutex - A Practical Distributed Mutual Exclusion...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online