Slashdot Mirror
Facebook Open Sources Its Network Routing Platform Open/R (techcrunch.com)
Facebook will open source its modular network routing software Open/R, currently used in its backbone and data center networks, which "provides a platform to disseminate state across the network and allows new applications to be built on top of it." An anonymous reader quotes TechCrunch:
Facebook obviously has unique scale needs when it comes to running a network. It has billions of users doing real-time messaging and streaming content at a constant clip. As with so many things, Facebook found that running the network traffic using traditional protocols had its limits and it needed a new way to route traffic that didn't rely on the protocols of the past, Omar Baldonado, Engineering Director at Facebook explained... While it was originally developed for Facebook's Terragraph wireless backhaul network, the company soon recognized it could work on other networks too including the Facebook network backbone, and even in the middle of Facebook network, he said. Given the company's extreme traffic requirements where the conditions were changing so rapidly and was at such scale, they needed a new way to route traffic on the network. "We wanted to find per application, the best path, taking into account dynamic traffic conditions throughout the network," Baldonado said.
But Facebook also recognized that it could only take this so far internally, and if they could work with partners and other network operators and hardware manufacturers, they could extend the capabilities of this tool. They are in fact working with other companies in this endeavor including Juniper and Arista networks, but by open sourcing the software, it allows developers to do things with it that Facebook might not have considered, and their engineering team finds that prospect both exciting and valuable.
"Most protocols were initially designed based on constrained hardware and software environment assumptions from decades ago," Facebook said in its announcement. "To continue delivering rich, real-time, and highly engaging user experiences over networks, it's important to accelerate innovation in the routing domain."
But Facebook also recognized that it could only take this so far internally, and if they could work with partners and other network operators and hardware manufacturers, they could extend the capabilities of this tool. They are in fact working with other companies in this endeavor including Juniper and Arista networks, but by open sourcing the software, it allows developers to do things with it that Facebook might not have considered, and their engineering team finds that prospect both exciting and valuable.
"Most protocols were initially designed based on constrained hardware and software environment assumptions from decades ago," Facebook said in its announcement. "To continue delivering rich, real-time, and highly engaging user experiences over networks, it's important to accelerate innovation in the routing domain."
For Facebook would be to /dev/null
I've never used facebook, blocked it at my router as well as all their tracking subdomains.
However I feel it's not enough. It offends me that facebook even exists, and that the average internet user is of the mentality to use it as their main form of communication.
How do we kill facebook?
'cos y'know standards...
> Facebook's value is the number of people signed up to it.
True
> Facebook will remain dominant simply by having the users to create enough content to be interesting. No upstart can beat that ...
Myspace was THE social networking site. Then Facebook challenged Myspace, then beat them. Network effects are.powerful, but not insurmountable.
One of the the more likely scenarios for something else to beat out Facebook is if they start in a very specific niche, such as DoctorNet, or CopBook . The service would be tailored to doctors, but could offer all the same features as Facebook. Once DoctorNet is strong among doctors, they'd add NurseNet, then RxNet, etc, growing their own network effects within a specific community and extending outward until they could challenge Facebook in general. Language specific networks could be another avenue. If a company had a strong presence in German, and Dutch, perhaps, they could add languages and cultures, then carefully interweave the different networks to create a super-network that retains specific communities as parts. Certain companies who run thousands of interest-specific web forums might be able o pull this off.
You can tell most of the geeks have left Slashdot when a technical story only get's about fifteen replies, because there's few left who understand the subject.
I have a low opinion of "intelligent cores", SDN, etc. No matter what the capabilities of physical interfaces and links are what really matters. You can't dynamically reconfigure reality in software to suite your administrative desires.
The result is systems with increasingly less and less purchase on physical reality and commensurate increases in complexity and overhead.
The most common reason people pull this shit is laziness. They don't want to bother creating systems or applications with the necessary feature sets to accomplish the job so they set off to offload responsibility to the network. It's one gross hack after another. The path of least resistance. The road to hell.
Stop making FB sound like it's rocket science. It's not.
Yeah Myspace had 80 million users alright - when Facebook had about half as much. Myspace had 50-60 million when Facebook had ZERO.
These figures show Myspace at 110 million, Facebook at 60 million:
http://www.web-strategist.com/...
> How is that likely? Because you said so? Is there really any evidence that people have any reason to flock to niche-oriented social networks?
Have you ever heard of a company called Facebook? How about Slashdot? The Slashdot effect, or "getting Slashdotted". Do you know why we don't read the articles before commenting? When Facebook had less than 1% the number of users as Myspace, they had nearly 100% of Harvard students. When they had half as many as Myspace, they had 85% of university students. That's how Facebook took on, and beat, the #1 social networking site - by getting a very specific niche, Harvard students, then growing that to university students, then twenty-somethings.
When I started participating on Slashdot (daily), most of the time you couldn't read the linked article because the site would be down from the swarm of other Slashdot users trying to read it at the same time. So many people would "flock to niche-oriented" sites to have conversations that the news-for-nerds niche site (Slashdot) had more users than most sites could handle. When a site was taken down by the sheer number of nerds of Slashdot, that was called "getting Slashdotted".
Theorizing certain workloads I can see where it would be beneficial to be able manage stp rules directly, especially seeing as how slow most switches react to changes and that the changes may be based on outside factors.
Managing routing from applications however is more common and has always been a patching affair since the industry has always taught "why rewrite what is already built" while it has been built with the idea "these are the settings that are good enough for most general cases".
Ive had to write code that manages private networks in cluster environments for tighter control and more predictable events. I hope this fb announcement brings networking manufacturers to expose their protocols to coders with deeper access APIs.
"Billions of users?"
Yeah, you're going to need to cite that assertion.
Kriston
It seems like the point of this is to find an improvement over the major network equipment manufacturers.
The announcement is really two things.
First a growable topology for hooking simple switches together.
Second the open, distributed software to control them to work together to make a functioning router.
The s/w is an interesting mix of use and don't use existing methods.
SDN, but not SDN protocols and implementations.
Link state ideas but in a more general way without the constraints of existing protocols.
The result is kind of like exploding a big iron router into an incrementally scalable set of hardware and server based control software.
This makes dealing with the router less of an expensive, locked in, mystical experience.
Instead of a big box full of switching gear with a back or mid plane connecting them together, the gear is in separate boxes with interconnecting cables.
This permits incrementally upgrading the thing by adding or swapping boxes and connections.
You don't have to get a forklift and stop and replace the big box as a whole.
You are not tied to a specific vendor for all of the boxes.
You are not tied to a specific vendor's rollout schedule.
If it works, seems like a good thing.
Others have probably already done this at this scale, but perhaps not with this degree of openness?