| SEDA: An Architecture for Highly Concurrent Server Applications Matt Welsh, Harvard University |
My Ph.D. thesis work at UC Berkeley focused on the development of a robust,high-performance platform for Internet services, called SEDA.The goal is to build a system capable of supporting massive concurrency(on the order of tens of thousands of simultaneous client connections)and avoid the pitfalls which arise with traditional thread andevent-based approaches.
SEDA is an acronym for staged event-driven architecture, anddecomposes a complex, event-driven application into a set ofstages connected by queues. This designavoids the high overhead associated with thread-based concurrencymodels, and decouples event and thread scheduling from applicationlogic. By performing admission control on eachevent queue, the service can be well-conditioned to load, preventingresources from being overcommitted when demand exceeds servicecapacity.SEDA employs dynamic control to automatically tune runtime parameters(such as the scheduling parameters of each stage), as well as tomanage load, for example, by performing adaptive load shedding.Decomposing services into a set of stages also enables modularity andcode reuse, as well as the development of debugging tools for complexevent-driven applications.
| February 19, 2007 - A Note on the status of SEDA I continue to receive many requests for information about SEDA. I am no longer actively working on this project, so all of these web pages should be regarded as "archival". It is also worth noting that a number of recent research papers have demonstrated that the SEDA prototype (in Java) performs poorly compared to threaded or event-based systems implemented in C. This would seem to contradict the findings in my work. (For more information I invite you to read recent papers by Vivek Pai's group at Princeton and the Capriccio work from UC Berkeley.) While I do not discount these later results, it is important to keep a few things in mind when interpreting them. First, the SEDA implementation in Java was developed and tuned on a particular JVM implementation (IBM JDK 1.3), on a particular version of the Linux kernel (2.2), using the /dev/poll event dispatch mechanism. More recent studies have varied the environment substantially. I have several theories about what could be causing this poor performance, although I have not had an opportunity to perform new measurements. I have noticed that performance of the SEDA networking layer is highly dependent on a number of parameters, such as the poll interval used by the various threads. This likely needs to be tuned or redesigned to support high bandwidth networks and more recent Linux and JVM implementations. Also, SEDA imposes a high context switch overhead in certain cases, depending on the number of threads and stages used, and the processing granularity within each stage. Tim Brecht's group at Waterloo has undertaken a study of competing Web server architectures and has shown that a SEDA implementation in C++, appropriately tuned, performs comparably to alternatives, so I do not believe these performance issues are fundamental to the architecture. The most fundamental aspect of the SEDA architecture is the programming model that supports stage-level backpressure and load management. Our goal was never to show that SEDA outperforms other server designs, but rather that acceptable performance can be achieved while providing a disciplined apporach to overload management. Clearly more work is needed to show that this goal can be met in a range of application and server environments. Please feel free to get in touch if you have new results or questions about the SEDA approach. |
Our current prototype of a SEDA-based services platform is calledSandstorm. Sandstorm is implemented entirely in Java and usesthe NBIOpackage to provide nonblocking I/O support. Support for the JDK 1.4java.nio package is included as well. Despite using Java,we have achieved performance that rivals (and sometimes exceeds) that ofC/C++. We have also implemented a SEDA-based asynchronous SSL and TLSprotocol library, called aTLS. All of this software is availablefor download below.
We have built a number of applications to demonstrate the SEDAframework. Haboob is a a high-performance Web serverincluding support for both static and dynamic pages.Other applications include a Gnutella packet router andArashi, a Web-based email service similar to Yahoo! Mail.
The best place to start for more information is theSOSP'01 paper on SEDA andthe corresponding talkslides.My Ph.D. thesis hasmuch more information as well.If you have questions, comments, or are interestedin collaborations, please feel free to contact me by e-mail(see my home page).
A number of open source and commercial systems are based on SEDA andNBIO. These include:
July 12, 2002: Lots of updates. CVS, release, and mailing listhosting is now athttp://seda.sourceforge.net.Now you can access the latest SEDA codebase via anonymous CVS,hopefully encouraging more collaborative development of the code.
The seda-users mailing list is back up - pleasesubscribe.
All of the code has been consolidated into a single CVS tree under thepackage name seda (renamed from mdw). The Haboob Webserver and aTLS code are also released and more completely documented.And a nice one-line performance patch to NBIO is included that increasesnetwork bandwidth by 30% or so!
File downloads are hosted by SourceForge.net.Click here for the SEDASourceForge Page. You may either download the SEDA software as a set ofpre-packaged "official" releases (.tar.gz format, source code included),or use anonymous CVS to access the "live" tree.The CVS tree will be updated more frequently than the"official" releases, which are meant to represent stable, testedversions of the software. The CVS tree is the "live code" that isunder constant development.
See the file README in each release for information oncompilation and usage. All of the SEDA code is covered underan open-source license (see below).
Official releases
All files are available from this SourceForge page. Or, you may click on one ofthe links below:
| Package | Latest version | Download |
|
| ||
| Latest SEDA "core" release: Includes the NBIO library and the Sandstorm runtime environment. | v3.0, July 12, 2002 | seda-release-20020712.tar.gz |
|
| ||
| Latest NBIO-only release: Includes only the NBIO library. | v2.0, July 12, 2002 | nbio-release-20020711.tar.gz |
|
| ||
| Haboob: A high-performance Web server built using Sandstorm. Requires the SEDA "core" release. | July 12, 2002 | haboob-release-20020712.tar.gz |
|
| ||
| aTLS: An asynchronous TLS and SSL protocol library for Sandstorm. Requires the SEDA "core" release. | July 12, 2002 | atls-release-20020712.tar.gz |
Anonymous CVS access
Anonymous CVS access is available for those users who want tomaintain a "live" source tree. To check out the SEDA tree usinganonymous CVS, use the following commands:
cvs -d:pserver:anonymous@cvs.seda.sourceforge.net:/cvsroot/seda login( Just press enter when asked for a password )cvs -z3 -d:pserver:anonymous@cvs.seda.sourceforge.net:/cvsroot/seda co seda
Javadoc API documentation
You can browse theJavadoc documentation for SEDA.
By transitioning the SEDA project to SourceForge.net, it is now possibleto open up the development of the SEDA code a wider community.If you are an active user of the SEDA or NBIO code and would like tocontribute, pleasejoin the seda-users mailing list.SourceForge makes it possible for all of the developers to share asingle CVS tree, make code releases, and so forth. I encourageinterested developers to join the team!
The best place to look for performance information about SEDA is thevarious papers about the system.Earlier (and somewhat outdated) performance results arediscussed in the following web pages:
The SEDA release is covered under the following Open Source license:
Copyright (c) 2002 by Matt Welsh and The Regents of the University ofCalifornia. All rights reserved.
Permission to use, copy, modify, and distribute this software and itsdocumentation for any purpose, without fee, and without written agreement ishereby granted, provided that the above copyright notice and the followingtwo paragraphs appear in all copies of this software.
IN NO EVENT SHALL THE UNIVERSITY OF CALIFORNIA BE LIABLE TO ANY PARTY FORDIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES ARISING OUTOF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN IF THE UNIVERSITY OFCALIFORNIA HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
THE UNIVERSITY OF CALIFORNIA SPECIFICALLY DISCLAIMS ANY WARRANTIES,INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITYAND FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER ISON AN "AS IS" BASIS, AND THE UNIVERSITY OF CALIFORNIA HAS NO OBLIGATION TOPROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
If you have any questions, comments, or bug reports, don't hesitate to getin touch with me!
联系客服
