I have a scientific program written in F# which I want to parallelize and run on 1 server with multiple processors (64) and for the future also in the cloud (Windows Azure?). The program will have a simple 1-1 communication between the nodes (no broadcast etc.).
If I used WCF, would it be as fast as MPI? What has MPI that WCF does not? There exists Pure MPI .NET written for WCF which puzzles me even more. I do not know if to use WCF or MPI.NET or Pure Mpi running on WCF.
PS: I guess that TPL is out of the game for 64 processors and more, right?
It is difficult to give a concrete answer, because it all depends on the specific aspects of your application, its current architecture (I suppose you already have some app) etc.
As you mention MPI and WCF, I assume that the application is written as several components that communicate with each other. The best way to structure this kind of application is to use F# agents.
As far as I understand, you want to run the application on a single server first. If you write it using agents, the agents can just communicate directly with each other (so you don't need MPI or WCF).
TPL should work well on a single-server (with lots of CPUs), but it will not scale to the distributed setting - you cannot run Task on another machine. However, you can use it inside individual components (e.g. agents) that will be distributed.
Regarding MPI vs. WCF - I don't have enough experience to answer that. However, if you use agent-based architecture, it should be easy to try various options. You may also check out fracture and related projects, which aims to implement high-performance sockets for F# (and possibly distributed agents in the future).
If you're doing it on 1 server you could just execute one process and execute the code in parallel. That way you could share memory more easily and faster than doing it through messages like MPI and WCF. Although the overhead of communication might not be that much, depending on your problem + solution.
Also the changes to your code would be much less that way, F# can usually be turned into prallel code with little effort. Going to MPI/WCF would require you to rewrite large portions.
Googling for F# + parallel gives plenty useful info that you should read first, like this for a good start:
http://blogs.msdn.com/b/dsyme/archive/2010/01/09/async-and-parallel-design-patterns-in-f-parallelizing-cpu-and-i-o-computations.aspx
So on 1 server, I woudl use the parallel features of F#, it's designed to prallelize easily.
Later when you want to go for cloud, that would be turning it into cleint-server. That's a different problem then parallization. I would treat and solve them seperately.
On the MPI vs WCF. WCF is designed as a RPC technology, i.e. you call remote procedures and get answers. If you want to use it for parallel programming with separate processes, you would have to create the boilerplate code for that. (Keep track of subsribed clients etc.)
MPI was designed to run that kind of architecture and handles it much more easily. (the first process gets number 0 and is the master, the other are slaves get numbered incrementally etc.)
Howver I don't think MPI will be very good to go cloud, since that invloves http, protocols, security etc. Not sure how well MPI works for those kind of things, WCF will handle that very well indeed.
The fact that there is an MPI.NET for WCF is because MPI is about a certain style of parallizing code that a lot of people are familiar with. So you can use the programming concepts and use them on the .NET platform leveraging WCF for the communications.
Something else you might want to look into if you need to exchange a lot of data over the wire is protocol-buffers (see protobuf-net for instance). That can easily be combined with WCF for communication and is very lean in serializing structured data so you can send over the wire efficiently.
Gert-Jan
WCF and MPI are different concepts. WCF is like a person A asks a person B to do something where as MPI is like a person A creates clones of himself (all clone have same ability/logic) and then these clones work on specific parts of the problem to be solved and once done they combine their results.
So choosing between which one fits your specific application depends on the problem your application is trying to solve. It may even be a combination of both WCF and MPI. Where your client application asks the WCF to do some task and the WCF create clones of the "problem solver" using MPI and when the clone are done with solving the problem (in parallel) they return the aggregated result back to the WCF and then that result is sent to client application.
You might also want to take at the 'mbrace' product, which provides a cloud monad (http://blogs.msdn.com/b/dsyme/archive/2011/08/23/m-brace-f-in-the-cloud.aspx). It's still at a fairly early stage though. I'm no expert but it may be that you can run an mbrace-based solution as effectively a private cloud on your 64-processor setup. When you outgrow that, a move to Azure would be seamless.
Related
I'm looking at designing a commercial web site from scratch in the new year and I was planning on using NServiceBus. Other components would include RavenDB, ASP.NET MVC, Ninject, Bootstrap, etc.
My question is, if I build scalability in from the beginning, particularly if in the first 6 months I plan to run the site from a single server, would it be a foolish thing to use NServiceBus from the outset? Will I experience much of a choke-point by pushing everything through MSMQ rather than direct calls to methods in DLLs? Should NServiceBus only be added to mature systems, or systems that are intended to be deployed to more than one server?
While the fastest possible call would be a direct in-memory invocation, I wouldn't look at optimizing for latency of calls in most web scenarios. Instead, focus on what type of logic can be run asynchronously with respect to the user request. That will be most influential on your overall scalability.
NServiceBus provides a fairly clean programming model for asynchronous invocations that can later on be distributed across multiple processes and machines.
I'm looking at building a Cocoa application on the Mac with a back-end daemon process (really just a mostly-headless Cocoa app, probably), along with 0 or more "client" applications running locally (although if possible I'd like to support remote clients as well; the remote clients would only ever be other Macs or iPhone OS devices).
The data being communicated will be fairly trivial, mostly just text and commands (which I guess can be represented as text anyway), and maybe the occasional small file (an image possibly).
I've looked at a few methods for doing this but I'm not sure which is "best" for the task at hand. Things I've considered:
Reading and writing to a file (…yes), very basic but not very scalable.
Pure sockets (I have no experience with sockets but I seem to think I can use them to send data locally and over a network. Though it seems cumbersome if doing everything in Cocoa
Distributed Objects: seems rather inelegant for a task like this
NSConnection: I can't really figure out what this class even does, but I've read of it in some IPC search results
I'm sure there are things I'm missing, but I was surprised to find a lack of resources on this topic.
I am currently looking into the same questions. For me the possibility of adding Windows clients later makes the situation more complicated; in your case the answer seems to be simpler.
About the options you have considered:
Control files: While it is possible to communicate via control files, you have to keep in mind that the files need to be communicated via a network file system among the machines involved. So the network file system serves as an abstraction of the actual network infrastructure, but does not offer the full power and flexibility the network normally has. Implementation: Practically, you will need to have at least two files for each pair of client/servers: a file the server uses to send a request to the client(s) and a file for the responses. If each process can communicate both ways, you need to duplicate this. Furthermore, both the client(s) and the server(s) work on a "pull" basis, i.e., they need to revisit the control files frequently and see if something new has been delivered.
The advantage of this solution is that it minimizes the need for learning new techniques. The big disadvantage is that it has huge demands on the program logic; a lot of things need to be taken care of by you (Will the files be written in one piece or can it happen that any party picks up inconsistent files? How frequently should checks be implemented? Do I need to worry about the file system, like caching, etc? Can I add encryption later without toying around with things outside of my program code? ...)
If portability was an issue (which, as far as I understood from your question is not the case) then this solution would be easy to port to different systems and even different programming languages. However, I don't know of any network files ystem for iPhone OS, but I am not familiar with this.
Sockets: The programming interface is certainly different; depending on your experience with socket programming it may mean that you have more work learning it first and debugging it later. Implementation: Practically, you will need a similar logic as before, i.e., client(s) and server(s) communicating via the network. A definite plus of this approach is that the processes can work on a "push" basis, i.e., they can listen on a socket until a message arrives which is superior to checking control files regularly. Network corruption and inconsistencies are also not your concern. Furthermore, you (may) have more control over the way the connections are established rather than relying on things outside of your program's control (again, this is important if you decide to add encryption later on).
The advantage is that a lot of things are taken off your shoulders that would bother an implementation in 1. The disadvantage is that you still need to change your program logic substantially in order to make sure that you send and receive the correct information (file types etc.).
In my experience portability (i.e., ease of transitioning to different systems and even programming languages) is very good since anything even remotely compatible to POSIX works.
[EDIT: In particular, as soon as you communicate binary numbers endianess becomes an issue and you have to take care of this problem manually - this is a common (!) special case of the "correct information" issue I mentioned above. It will bite you e.g. when you have a PowerPC talking to an Intel Mac. This special case disappears with the solution 3.+4. together will all of the other "correct information" issues.]
+4. Distributed objects: The NSProxy class cluster is used to implement distributed objects. NSConnection is responsible for setting up remote connections as a prerequisite for sending information around, so once you understand how to use this system, you also understand distributed objects. ;^)
The idea is that your high-level program logic does not need to be changed (i.e., your objects communicate via messages and receive results and the messages together with the return types are identical to what you are used to from your local implementation) without having to bother about the particulars of the network infrastructure. Well, at least in theory. Implementation: I am also working on this right now, so my understanding is still limited. As far as I understand, you do need to setup a certain structure, i.e., you still have to decide which processes (local and/or remote) can receive which messages; this is what NSConnection does. At this point, you implicitly define a client/server architecture, but you do not need to worry about the problems mentioned in 2.
There is an introduction with two explicit examples at the Gnustep project server; it illustrates how the technology works and is a good starting point for experimenting:
http://www.gnustep.org/resources/documentation/Developer/Base/ProgrammingManual/manual_7.html
Unfortunately, the disadvantages are a total loss of compatibility (although you will still do fine with the setup you mentioned of Macs and iPhone/iPad only) with other systems and loss of portability to other languages. Gnustep with Objective-C is at best code-compatible, but there is no way to communicate between Gnustep and Cocoa, see my edit to question number 2 here: CORBA on Mac OS X (Cocoa)
[EDIT: I just came across another piece of information that I was unaware of. While I have checked that NSProxy is available on the iPhone, I did not check whether the other parts of the distributed objects mechanism are. According to this link: http://www.cocoabuilder.com/archive/cocoa/224358-big-picture-relationships-between-nsconnection-nsinputstream-nsoutputstream-etc.html (search the page for the phrase "iPhone OS") they are not. This would exclude this solution if you demand to use iPhone/iPad at this moment.]
So to conclude, there is a trade-off between effort of learning (and implementing and debugging) new technologies on the one hand and hand-coding lower-level communication logic on the other. While the distributed object approach takes most load of your shoulders and incurs the smallest changes in program logic, it is the hardest to learn and also (unfortunately) the least portable.
Disclaimer: Distributed Objects are not available on iPhone.
Why do you find distributed objects inelegant? They sounds like a good match here:
transparent marshalling of fundamental types and Objective-C classes
it doesn't really matter wether clients are local or remote
not much additional work for Cocoa-based applications
The documentation might make it sound like more work then it actually is, but all you basically have to do is to use protocols cleanly and export, or respectively connect to, the servers root object.
The rest should happen automagically behind the scenes for you in the given scenario.
We are using ThoMoNetworking and it works fine and is fast to setup. Basically it allows you to send NSCoding compliant objects in the local network, but of course also works if client and server are on he same machine. As a wrapper around the foundation classes it takes care of pairing, reconnections, etc..
I have decided to use Twisted for a project and have developed a server that can push data to clients on other computers. At the moment I am using dummy data for testing speed requirements but I now need to interface Twisted to my other Python DAQ application which basically collects real-time data (500 Hz) from various external devices over different transports (e.g. Bluetooth). (note: the DAQ (data acquisition) application is on the same computer as the Twisted server)
Since the DAQ application is not part of the Twisted framework I am wondering what is the most efficient (fastest, robust, minimal latency) way to pass the data to the Twisted server. I have considered using a light-weight database, memcache, Queue or even the Twisted plugins but it is hard to tell which would be the most appropriate and best fit. I should add that the DAQ application was developed before deciding on using Twisted so I have so far considered it as separate from the Twisted network.
On the other side of the system, the client side, which reside on multiple computers, I have a similar problem. As the data streams in (I am sending lines of data, about 100 bytes each) I want to hand this data off to another application which will process this data for a web application (I would prefer to use Twisted Web Service for this but that is not my choice!) The Web application is being written in Java. Once again I have considered the choices above but since I am new to Twisted I am not sure which is the best approach. (note: the Web application is on the same computers as the Twisted clients)
Any advice or thoughts would be greatly appreciated.
My suggestion would be to to build a simple protocol with twisted's built-in support for AMP; you can hook this in to any other languages or frameworks using one of the implementations of AMP in other languages. AMP is designed to be as easy as possible to implement, as it's just a socket with some length-prefixed strings arranged into key/value pairs.
There's obviously a zillion different ways you could go about this, but I would first look at using a queue to pass the data to your Twisted server. If you deploy one of the many opensource queueing tools (e.g. RabbitMQ, ZeroMQ, OpenMQ, and loads of others), you should be able to write from your DAQ product using something generic like HTTP, then read into your Twisted server also using HTTP. If you don't like HTTP, then there would be a lot of alternative transports to choose from - just identify which you want to use, then use that as a basis for selecting your queueing tool.
This would give you an extremely flexible solution, in that you could upgrade or change any of these products with minimal impact to anything else in the whole solution.
I work for a large state government agency that is a tad behind the times. Our skill sets are outdated and budgetary freezes prevent any training or hiring of new employees/consultants (firing people is also impossible). Designing business objects, implementing design patterns, establishing code libraries and services, unit testing, source control, etc. are all things that you will not find being done here. We are as much of a 0 on the Joel Test as you can possibly get. The good news is that we can only go up from here!
We develop desktop CRUD applications (in C++, C#, or Java) that hit the Oracle database directly through an ODBC connection. We basically have GUI's littered with SQL statements and patchwork code. We have been told to move towards a service-oriented n-tier architecture to prevent direct access to the database and remove the Oracle Client need on user machines.
Is WCF the path we should be headed down? We've done a few of the n-tier application walkthroughs (like this one) and they seem easy to implement, but we just don't know enough to understand if we are even considering the right technologies. Utilizing the .NET generated typed DataSets seems like a nice stopgap to save us month/years of work (as opposed to creating new business objects from the ground up for numerous projects). Is this canned approach viable for a first step?
I recently started using WCF services for my Data Layer in some web applications and I must say, it's frustrating at the beginning (the first week or so), but it is totally worth it once the code is deployed.
You should first try it out with a small existing app, or maybe a proof of concept to make sure it will fit your needs.
From the description of the environment you are in, I'm sure you'll realize the benefit almost immediately.
The last company I worked for chose WCF for almost the exact reason you describe above. There is lots of good documentation and books for WCF, its relatively easy to get working, and WCF supports a lot of configuration options.
There can be some headaches when you start trying to bend WCF to work in a way not specifically designed out of the box. These are generally configuration issues. But sites like this or IDesign can help you through those.
First of all, I would definitely not (sorry for the emphasis) worry about the time you'll save using typed DataSet's versus creating your own business objects. That is usually not where you will spend most of your development time. I prefer using business objects myself.
In you're situation I would want to implement a proof-of-concept first. One that addresses all issues you may encounter. This proof-of-concept should implement an entire use case, starting on the client, retrieving data from the database and returning it to the client. You should feel confident about your implementation before continuing.
Then about choice of technology. WCF is definitely a good choice for communication between your client applications and the service layer. I suppose that both your clients as well as your service layer will become C# applications? That makes things a lot easier since interoperability between different platforms (Java/C# for example) is still not trivial although it should work in most cases.
Take a look at Entity Framework (as there are a couple Oracle providers available for it already) in conjunction with .NET 3.5 SP1 which enables built-in WCF serialization of your EF generated classes.
Here is a good blog to get started: http://blogs.msdn.com/dsimmons
CSLA might be a good fit for your N-Tier desktop apps. It supports WCF, has a large dev community, and is well documented. It is very object oriented.
In embedded systems using linux for an application; under which condition will you divide an application into two/three processes. My main doubt is; is it required to divide a single application component into multiple processes and then run multiple processes to achieve the required application functionality.
By experience I tend to isolate possibly problematic pieces of code. For example if you're depending on a sensor which ships with 3rd party libraries that you do not trust, making it a separate process will make your application more robust and fault-tolerant because you'll be (hopefully) able to restarts only parts of it.
Also for integration purposes it might be easier. Suppose your process A runs fine, then you can plug in the process B easily instead of adding new parts to process A. It might not seem like a big plus right now but it depends a lot on your project.
It does come with some overhead however, as dealing with synchronization and message passing can be more complicated and add to the design.
You don't have to do anything like that however.
You don't tell much about the circumstances that led you to your question, so I can only guess what kind of answer you are interested in.
Linux offers multi-threading functionality since ages, so concurrent programming can be done without multiple processes.
There is rarely a functional reason to divide integral components of an application into processes.
My suggestion is to write a single-process application. Should the requirement arise, that is: a problem can only be solved by managing runtime resources in separate processes, you can still take on the heavy work of solving inter-process communication and resource sharing, without having to change much in your business logic.