MPI-INF Logo
Campus Event Calendar

Event Entry

New for: D1, D2, D3, D4, D5

What and Who

Scaling TCP performance for multicore systems

KyoungSoo Park
KAIST
SWS Colloquium

KyoungSoo Park is an associate professor in the Electrical Engineering department at KAIST. His research interests focus on the reliability, performance, and security issues in the design and implementation of networked computing systems. He has developed CoBlitz, a scalable large-file content distribution network (CDN), which is acquired by Verivue, Inc., and subsequently by Akamai, Inc. He has co-developed HashCache, a memory-efficient caching storage system for developing regions, which was chosen one of the top 10 technologies in 2009 by the MIT technology review magazine. Most recently, his mTCP paper received the community award at USENIX NSDI 2014. He received his B.S. degree from Seoul National University in 1997, and his M.A. and Ph.D. degrees from Princeton University in 2004 and 2007, respectively, all in computer science. Before joining KAIST, he worked as assistant professor in the computer science department at University of Pittsburgh in the year of 2009.

AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, MMCI  
Expert Audience
English

Date, Time and Location

Tuesday, 9 December 2014
11:30
60 Minutes
E1 5
029
Saarbrücken

Abstract

Scaling the performance of short TCP connections on multicore systems is fundamentally challenging. Despite many proposals that have attempted to address various shortcomings, inefficiency of the kernel implementation still persists. For example, even state-of-the-art designs spend 70% to 80% of CPU cycles in handling TCP connections in the kernel, leaving only small room for innovation in the user-level program.

 In this talk, I will present mTCP, a high-performance user-level TCP stack for multicore systems. mTCP addresses the inefficiencies from the ground up - from packet I/O and TCP connection management to the application interface. In addition to adopting well-known techniques, our design (1) translates multiple expensive system calls into a single shared memory reference, (2) allows efficient flow-level event aggregation, and (3) performs batched packet I/O for high I/O efficiency. Our evaluations on an 8-core machine showed that mTCP improves the performance of small message transactions by a factor of 25 compared to the latest Linux TCP stack and a factor of 3 compared to the MegaPipe system. It also improves the performance of various popular applications by 33% to 320% compared to those on the Linux stack.

Contact

Brigitta Hansen
0681 93039102
--email hidden

Video Broadcast

Yes
Kaiserslautern
G26
113
passcode not visible
logged in users only

Christian Klein, 10/13/2016 17:09
Brigitta Hansen, 12/04/2014 16:18 -- Created document.