Cisco To Open-Source New Messaging Protocol

Esther Schindler writes "Do you use SOAP, CORBA or EJBs? You might want to take a look at Etch, writes James Turner for CIO.com. It's language-, platform- and transport-agnostic, and Cisco is planning to release it as open source. Certainly, it offers some technical benefits: 'In addition to a simplified configuration, Etch also promises less overhead over the wire, compared to SOAP. In a testbed environment where SOAP was managing around 900 calls a second, Etch generated more than 50,000 messages in a one-way mode, and 15,000 transactions with a full round-trip, company officials stated.' And the open source part? Cisco is in the process of deciding what license to use. 'The intent is to use a less restrictive license than GPL, perhaps Apache or Mozilla. This is to allow commercial developers to incorporate Etch into products without licensing issues. A final announcement on the licensing decision will be available in the next month.'"

Comments

Distributed computing.

[William Robinson]

Having developed many solutions using CORBA and SOAP, I welcome better solutions. Not sure though how it is going to resolve the problems we faced with CORBA and SOAP both?

SOAP could be easily integrated over current HTML based networks without need to make hole in firewall. But it was pig slow and designing stateful services was painful.

CORBA offered more technical challenges viz, complexity, version control, fault tolerance (not that a SOAP HA services is piece of cake, but I don't want anybody to go through torture of designing a HA CORBA server)

ZeroC's ICE

[bheer]

It'll be interesting to compare Etch to ICE, which is a GPL'd open-source, cross-language RPC toolkit (you can buy commerical licenses too). It's quite widely used by banks and is generally reckoned to be speedy.

CORBA was shit. SOAP not much better.

[Anonymous Coward]

All of these distributed technologies have been shit. CORBA was absolutely hell to develop with. Besides the runtime performance problems, development was always a huge hassle. It rarely just worked. J

Java's RMI was slightly better. But again, the development overhead was huge. Generating proxy and stub classes becomes a chore really quickly, and debugging becomes a real challenge.

SOAP was a little bit better than CORBA and Java RMI. At least writing the object layer code is a far more reasonable task. The performance, though, was complete shit compared to Java RMI and Corba. Whatever development time you saved initially in writing the SOAP interfacing code was instead spent trying to optimize what you had so that it wouldn't perform so fucking horribly.

In some ways, I hope that Cisco can do better. But I really don't know if that's possible. It may just be the nature of the beast that these sort of technologies perform poorly, are slow to develop, and are often nothing more than a huge hassle.

Re:CORBA was shit. SOAP not much better.

[RAMMS EIN]

I think the problem with all of these is that they consist of too much hype and too little sense. Also, a lot of them are horribly over-engineered.

Taking SOAP as an example, because that's the one I know best. What you need is some way to communicate data to another party. What you get is something that does that, but in about the most verbose and latency-sensitive way imaginable. Yes, it's standards-based, which is a Good Thing, and it's human-readable, which is also advantageous.

On the other hand, the standard is apparently complex enough that various implementations implement distinct subsets...meaning interoperability is hit and miss. And seriously...HTTP and XML? Horribly inefficient.

All I know about Java RMI is that it was a horrible pain to get everything set up right so that it would work...and even then, of course, it's _Java_ RMI, so I'm not sure it's the best choice if you don't want to prescribe a particular platform or object model.

What I think all these have in common is that they try to automagically solve a problem that can hardly be solved automagically. The premise, as far as I understand, of protocols like this is that they will let you exchange objects from your program with programs on remote machines, without you having to do any work on it. But, in practice, it's hard to decide how to go about that without specific knowledge of how these objects will be used. Specifically, objects tend to be part of a graph...which parts of the graph do you send over the wire? And then there's all the actually hard problems related to distributed programming...

As for the protocols, I've always thought it shouldn't be that hard. Lisp has this thing where you can read and write Lisp objects. Write it out and read it back in and you'll get an equal object. This is incredibly useful for many things, including debugging, but also makes it easy to envision a protocol for distributed computing - especially when you consider that Lisp programs also consist of Lisp objects.

On the other hand, Lisp's read-write protocol isn't the most efficient as far as parsing and transmission are concerned. So something could be gained there. And not all objects can be read back in - for some it would even be a Bad Thing if they could. And none of the harder problems (like which parts of the graph to send) have been addressed yet. On the gripping hand, I honestly feel it's already better than SOAP.

For the rest, I think the trick is to keep things simple. Think of what you want to send. Encode that using some simple, efficient encoding, and send it. Don't shoot for automagic, Just Works exchange of arbitrary program objects. Think about what you really need to send and send only that.

Oh, and don't draw in layers and layers of frameworks, please.

If this catches on then how long would it be

[Anonymous Coward]

before Microsoft releases their own just ever so slightly different version that was totally incompatible with any other version and made all its own tools work with their version only?

They will no doubt trumpet loudly their 'innovation' at the same time.

I hope Cicso license this in such a way that they could stop this sort of trick that M$ has played before on an 'open standard'

Cisco, Please use the LGPLv3 license.

[NZheretic]

Releasing the implementation library under a LGPL license will still allow for the functionality to be incorporated ( via dynamic linking ) into any proprietary product. The LGPL will insure the availability of the source code and downstream legal reuse rights of Cisco's implementation to downstream recipients.

The LGPL is the only license that will insure that at least that Cisco's implementation of the protocol can not be easily extended in an inoperative manner.

Given the timespan that Cisco expects the protocol to be in use, version 3 of the LGPL is the best option.

Ice?

[IamTheRealMike]

Other than license, how does this compare to ZeroCs Iceï¼Y Does anybody know? I've played with Ice before and it's very well done, although I remain to be convinced of the value of remote object references in a distributed system.

Imagination

[CarpetShark]

More and more companies moving away from GPL? That's a strange conclusion, considering that it's probably had the fastest growing mindshare an uptake of any software license, ever, and that GPLv3 is proving very popular already with new projects and migrations.

There's absolutely no ethical reason to choose a less restrictive license over the GPL. The only thing the GPL restricts is the ability to restrict others. THAT is possibly a reason to avoid it, since, for example, I would like to prevent military types from using things I worked on, but avoiding the GPL because you want corporations to have the ability to use public works it in works they then keep from the public is a VERY strange notion.

Re:Imagination

[Timothy Brownawell]

There's absolutely no ethical reason to choose a less restrictive license over the GPL. That depends on who you ask.

The only thing the GPL restricts is the ability to restrict others. Funny thing is, it isn't possible anyway to "restrict others" in that fashion without their cooperation (buying/downloading your software). It restricts the choices available to the end user by causing certain products to not exist.

Re:Microsoft's Kerberos

[Timothy Brownawell]

"Here's the spec, do whatever you want with it, but you can only use our name for it if you pass this huge test suite."

Re:GPL

[ruin20]

I've never met a GPL code developer who released his code under GPL because he was forced. I support GPL because I believe if something is important it should be codified and that if you develop something for the community you should protect it for the community. But that doesn't mean that releasing something under an FOSS license without a "recontribute/openness" clause doesn't mean that there won't be active community development. Something built on and from sharing will always foster more sharing, it's an issue of principal.

Re:GPL

[Ian Alexander]

Sorry to feed the troll, but the point of the GPL is not to increase adoption. Your absolutely right to say that other licenses will lead to greater adoption- but this is adoption by people who may take, take, take and not give back.

The company I work for sells closed source software. We also use some open source software (not GPL) in the product.

We contribute back to the open source we use because it's more sensible. Adding the same features back in again and again would be counterproductive. We'd rather they get added to the open source project permanently.

We have a blanket ban on using GPL'd source, though. We can't afford to GPL our entire 20 million line software stack, which would be the result of using even a tiny bit of GPL code.

Try to understand that not everyone loves the GPL and not everyone that doesn't love the GPL is a troll.

Now it's my turn to get modded into oblivion for not being fond of the GPL. Sigh.

Is that really so? I thought the GPL only "infected" if you had to link against the GPL'ed code. Or is your codebase that... interconnected? And what about LGPL? Inquiring minds want to know!