I've used XMPP extensively for several projects, from python, JavaScript, and Squeak - the wonderful thing about it is that you can start off simple using just chat messages, but then add more features like pub sub, conference rooms, service discovery, etc. Scalability is built into the system, and you start out with Prosody jabber server (easy to get going, written in Lua).
There are a lot of good books on the protocol and its supported by most environments you can think of... Now for the "bad" news. The library I used in smalltalk was written by Michael Rueger, and I don't think it has been updated for years; I think there is a version on squeaksource, so it would have to be brought up to current state of XML libs and so forth. Anyway, I like XMPP ;-) - Brian On Nov 19, 2014 1:29 PM, "Sven Van Caekenberghe" <s...@stfx.eu> wrote: > > > On 19 Nov 2014, at 20:52, Esteban A. Maringolo <emaring...@gmail.com> > wrote: > > > > My current use case is pretty simple, being able to push requests to the > remote devices, and gather the data back, assuming the remote devices > (clients) are already connected to the server, which is a client-related > problem with any messaging solution. > > > > I learn about MQTT when looking for alternatives to Google Cloud > Messaging for Android (which is built on to of XMPP). Facebook Messenger > uses MQTT. > > > > I don't want to develop an overkill solution that could delay > deliverance of value to my customer, though I find implementing an MQ/MQTT > solution intellectually challenging. I even read the standard a few months > ago, and it seems to be simpler than expected. I'd love to go down the > "implement MQTT server in Pharo", I would lose a real MQ scalability, but > it could be a good way to learn the protocol, and an alternative to > Mosquitto [1] with the benefits of having a server inside the image. > > > > Do you have experience with the scalability of Zinc WebSockets? > > How many of them can a single Pharo image handle? The I/O would be > minimal (< 1KB) per message, and the requests would be measured in minutes > instead of seconds. Although a 50 ws frames/sec peak during a couple of > seconds is feasible. > > I don't have any production level experience with (Zinc) WebSockets. But > we have some Pharo images in production doing related things. Based on that > I would say 100s of concurrent open connections is doable, but I would be > careful if you know you will scale to 1000s. Why ? Because each connection > is a socket(stream) with 3 semaphores and a process/thread. It will still > work, but it might be harder to keep things together. BTW, this is not a > Pharo problem per se, it will be hard(er) everywhere. > > I just read that MQTT over WebSockets also exists ;-) > > I guess that you have no RabbitMQ experience so maybe that scares you a > bit, you will have to learn some new things, but it is not hard (and it is > quite interesting). > > In terms of doing MQTT directly in Pharo, I would start with the client > side, it can't be very hard to do (there are many clients already). > > > Regards, > > > > [1] http://mosquitto.org/ > > > > El Wed Nov 19 2014 at 4:22:56 PM, Sven Van Caekenberghe <s...@stfx.eu> > escribió: > > Hey Esteban, > > > > Since you seem to need to scale to 1000s of concurrent connections, I > would go for option 2. RabbitMQ will then work as a demultiplexer and you > gain options for using more than one Pharo worker for processing traffic, > if that ever becomes necessary. > > > > BTW, I never heard of MQTT and it seems pretty interesting, thanks ! I > will read about it tonight. > > > > Sven > > > > > On 19 Nov 2014, at 19:06, Esteban A. Maringolo <emaring...@gmail.com> > wrote: > > > > > > Hi, > > > > > > I'll be needing to add bidirectional messaging to our current Android > (Java) application, and as usual I'm expecting to manage the server side of > the communication with Pharo. > > > > > > I'm evaluating two alternatives, but I'm in open to other, proven, > options. > > > > > > Option 1. WebSockets > > > Open a WebSocket on each device, and push/retrieve messages from each > of these, tracking the device id if I need to recreate a new socket > (avoiding recreating new sockets in case of connectivity issues, very > common). > > > > > > Option 2. MQTT [1] > > > This is basically a mobile oriented MQ, super low footprint on mobile. > > > For what I saw, I could implement the server using an intermediate MQ > (RabbitMQ) and use STOMP to connect to it. > > > > > > Option 3. XMPP [3] > > > Provides several features I'll need in the future, like file transfer > in addition to regular text messaging. It is very well supported in Android > with Smack [4], but I don't know if we have a Pharo server for it, or if > somebody ever played with it. > > > > > > By means of simplicity and use load I favor option 1, because I can > understand it better, and as everything moves towards web based > technologies, I could implement WAMP [5] on top of that in the future. > > > > > > Right now I'd need to have a hundred websockets opened at the same > time, which doesn't sound like a heavy load to me, but I certainly don't > know. In the future it could be an order of magnitude bigger, and that's > why I don't discard more complex solutions like XMPP or MQTT. > > > > > > Regards! > > > > > > > > > [1] http://mqtt.org/ > > > [2] https://github.com/svenvc/docs/blob/master/neo/stamp.md > > > [3] http://xmpp.org/ > > > [4] http://www.igniterealtime.org/projects/smack/ > > > [5] http://wamp.ws/ > > > > > > > > > > > > >
