The implementation includes a user space switch, a kernel space switch and NetFPGA implementation and is based on OpenFlow v0.89.

It supports MPLS by adding two tupels to standard OpenFlow tupels, which can match on up to two top of the stack MPLS labels. The rewrite action is supported in flow table for MPLS tags, but other MPLS related actions (Push, Pop, TTL and EXP bits operations) are supported via MPLS-enabled virtual ports.

Update:  A patch has been added as indigo-1.0-alpha-2.tgz.  This addresses a critical memory leak.

Indigo is the code name of the OpenFlow 1.0 software release for the Quanta LB4G, a 48-port Gigabit switch with 4 10 Gbps uplinks based on the Broadcom 56514.  The code targets other Broadcom devices as well, although this release supports only the LB4G.  The code is the follow-on to the Stanford-LB4G software release from last September.  Indigo differs from the previous release in several ways:

• Supports OpenFlow 1.0
• Linux user space code:  Only the minimal device drivers are installed in the kernel.  The majority of the code uses the OpenFlow udatapath implemenation.
• Hardware driver abstraction:  The code is written to a new hardware abstraction API that extends the current user space software table structure allowing easier porting to new hardware platforms.  The API header file and updated OpenFlow code is also available (see below).
• Targets other Broadcom devices including reference designs for the 56634 and 56820.

This is an alpha release and still has some known issues, but should provide a basis for experimentation.  Your feedback is strongly encouraged to help prioritize features and make this usable for OpenFlow deployments in the enterprise, data center, and wide-area network.

The initial release notes are available here.  A very preliminary hardware roadmap page is available here.

To receive access to the software or to provide feedback regarding the package, please send email to info@openflowswitch.org and include “LB4G” in the subject line.

Unless the winter weather stops us, we will within five days visit:

• Rutgers
• Princeton
• GeorgiaTech
• Clemson
• University of Indiana (Bloomington)
• University of Wisconsin at Madison
• Washington University

We hope to post a short update from each university while we are on the road, as well as an overall summary at the end. This should be interesting. Stay tuned.

Unlike previous releases, OpenFlow 1.0 is the first release of the standard that we feel is ready to be implemented in generally available products. We have been running OpenFlow in our Stanford network over the past months, and it has proven itself for both production use as well as experimentation.

OpenFlow 1.0 adds a number of key features. The largest addition is Slicing, a simple QoS mechanism that allows the isolation of traffic in OpenFlow networks. Smaller changes include matching IP addresses in ARP packets, Flow Cookies to identify flows, selective port statistics and matching on the ToS bits in the IP header. The release also includes a large number of small changes to the specification and bug fixes in the reference implementation. For a complete list of changes read the Release Notes or the more detailed Wiki Page.

Hardware accelerated OpenFlow 1.0 capable switches are expected to become available over the next months and we will keep you updated on the OpenFlow Blog and the OpenFlow-Announce Mailing List. On the controller side the reference controller supports 1.0 and there is a version of NOX with partial support available. If you deploy OpenFlow 1.0, feedback and bug reports are highly welcome via Trac or the openflow-discuss mailing list.

This release would not have been possible without the work of (in alphabetical order) Guido Appenzeller, Peter Balland, Martin Casado, David Erickson, Glen Gibb, Brandon Heller, Mikio Hara, Bob Lantz, Masayoshi Kobayashi, Nick McKeown, Justin Pettit, Ben Pfaff, Rob Sherwood, Srini Seetharaman, Dan Talayco, Jean Tourrilhes, Tatsuya Yabe, KK Yap, Yiannis Yiakoumis as well as to the many other members of the OpenFlow community that provided valuable suggestions, feedback and testing.

Thanks to all of you!

Brandon, Glen and Guido

The release candidates are being made available to provide the community with an opportunity to provide feedback prior to the official release. We plan on making the release official by the end of year (12/31) so please provide feedback before this date. Bugs reported in the reference software may not be corrected prior to the official release depending upon their severity; we plan to make a bug fix release in the new year that will address as many bugs as possible.

New features

New features added to the 1.0 specification include:

• Slicing support (multiple queues per port with minimum bandwidth guarantees)
• Matching on IP fields inside ARP packets
• Matching on IP ToS bits
• Improved flow duration resolution in stats/expiry
• Opaque flow cookies added to flows
• Ability to retrieve port stats for individual ports
• User-specifiable datapath description added to desc stats

The updated specification also includes numerous clarifications. For a more detailed list of changes please see the release notes (http://www.openflowswitch.org/wk/index.php/OpenFlow_1.0_release_notes) and the development wiki (http://www.openflowswitch.org/wk/index.php/OpenFlow_v1.0).

Reference switch changes

The reference switch is 1.0-RC compliant.

Kernel-mode support (comprising the datapath kernel module) has been dropped from the reference implementation. We plan a separate kernel module release some time in 1H2010.

Testing

This is the best-tested release that we’ve ever made available. The reference switch has been run in a production network for a period of two weeks to test the new features and general stability of the code base.

The test network was run under the control of two different controllers: the reference controller (distributed as part of the reference) and a 1.0-compatible version of NOX 0.6. Each controller was run for a period of about a week. Special testing modules were used with NOX to exercise each of the new features.

The reference switch has also successfully passed the regression tests on the following Linux distributions:

• Ubuntu 9.10 (32-bit and 64-bit)
• Debian unstable (32-bit)
• Fedora Core 12 (32-bit)
• CentOS 5.4 (32-bit)

Known issues

The current implementation of slicing has been observed to provide poor isolation between queues on the same port when run on systems with low-performance embedded CPUs. These problems have not been observed on systems with higher-performance CPUs.

Please see the release notes for other known issues.

Feedback and bug reporting

One important way to improve OpenFlow is to provide feedback and report bugs. Two mailing lists exist for discussing issues related to the spec and reference implementation. These lists are:

• spec: openflow-spec@lists.stanford.edu
• reference switch and development: openflow-dev@lists.stanford.edu

Please visit http://mailman.stanford.edu/ to subscribe to these lists.

Please report bugs using the OpenFlow Trac system: http://www.openflowswitch.org/bugs/openflow
Please provide as much information as possible when submitting bugs to aid the process of reproducing and diagnosing the bug.

Download

The specification and reference switch can be downloaded from:

• http://www.openflowswitch.org/documents/openflow-spec-v1.0.0-rc.pdf
• http://www.openflowswitch.org/downloads/openflow-1.0.0-rc.tar.gz

The specification and reference switch can also be obtained via git:

git clone git://openflowswitch.org/openflow.git
cd openflow
git checkout -b devel/1.0-rc origin/devel/1.0-rc

Acknowledgments

We would like to thank everyone who contributed to OpenFlow to help make this release possible.

It is split into four parts (clicking on the links below starts the video):

• Part1: Setting up a local OpenFlow Substarte (5 min)
• Part2: Managing Aggregates and Users (2.5 min)
• Part3: Protecting your Network (2 min)
• Part4: Creating a Slice and running Experiments (7 min)

Enjoy!

As a proof of concept, we built a simple OpenFlow enabled packet and circuit switch network using carrier-class Ciena CoreDirector CI  switches. The CDs natively support the OpenFlow protocol for their packet and circuit switching fabrics. We then built an application in the OpenFlow controller that sets up, modifies and tears down packet and circuit flows on demand and dynamically responds to network congestion. Learn more about the demo in the OFC’10 invited paper on the publications page. Our larger goals and motivations are detailed here.

Thanks to Dan Getachew, Preeti Singh, Lyndon Ong, Jim Archuleta, Puneet Gupta, Ashok Kumar, Ritesh Ralhan and the rest of the Ciena development team, as well as Yiannis Yiakoumis, KK Yap, and Vinesh Gudla from the Stanford side. Thanks also to  the Ciena Marketing and Sales teams for putting together the demo booths, material  and other support — it was good to see the buzz the demo created in the Research & Educational Networks (REN) community.  Photo Gallery: SC09

]]> http://www.openflowswitch.org/wp/2009/11/openflow-demo-at-sc09/feed/ 5 http://www.openflowswitch.org/wp/2009/11/openflow-demo-at-sc09/ http://feedproxy.google.com/~r/OpenFlowBlog/~3/kh74h7KjzkQ/ http://www.openflowswitch.org/wp/2009/11/openflow-demo-at-gec6/#comments Wed, 18 Nov 2009 02:07:18 +0000 Guido Appenzeller http://www.openflowswitch.org/wp/?p=1043 We had another major demo of OpenFlow today at the GENI Engineering Conference in Salt Lake City, Utah. The demo showed the use of a GENI Clearing House to reserve a virtual network slice across OpenFlow Networks at Stanford University, across the Internet2 Backbone and at Princeton University.

It was great to see that now a small number of projects inside GENI are making progress towards becoming deployable systems. Compliments to Rob Ricci’s demo of ProtoGENI which was very well done.

The evening before the major demo at the plenary, we also exhibited at the demo session that was held in the tower of the Rice Eccles Stadium. Thanks to Rob Sherwood, Srini Seetharaman, Jad Naous,  Masayoshi Kobayahsi, Sapan Bhatia, Tony Mack, and Andy Bavier for the hard work that it took to make this happen. It was worth it!

Photo Gallery: GEC6

The OpenRoads Team

n-casting between WiFi-WiMAX using OpenRoads is an example of our vision.  It was also part of SIGCOMM’s best demonstration.  An improved demonstration was shown in Mobicom and won honorable mention.  This demonstration has been shown in several other occasions, such as the OpenFlow workshop at Stanford.  If you have not yet seen the demonstration, a video of it is available on this website.

Now you are also able to reproduce this demonstration.  We are proud to make this demonstration available to the community.  This includes instructions for the physical setup and streaming server.  We have also made available the source code for the network controller, wireless client and visualization.  More importantly, we make OpenFlow Wireless available to public, to promote innovation in the wireless networks we use every day.

n-casting and OpenFlow Wireless are made available to you by the team pictured above, with the help of many others.