Utilizing amazon gpu cluster in web app - cuda

Does anyone have any experience plugging into an Amazon gpu cluster in a web app? Is it even possible? I want to make an online interface to a simulator I have written in cuda, but everything I have seen so far involves ssh and the command line.

Yes, this is very possible!
You may want to consider binding your CUDA C code (I'm just assuming C) to your preferred scripting language for better efficiency if you intend to build the front end logic not in a lower level programming language but absolutely this is possible.
Keep in mind, many web developers build their apps strictly from the command line, you simply need to find an effective and efficient way of calling the CUDA code from your Web app and then simply return the results to the client who made the request.

Related

Is it possible for an application to be built by only pure functions?

As the title says, regarding a part of functional programming I have yet to see anyone discuss or answer areas of applications which isn't possible to write pure functions. I feel that it is indeed impossible to write applications with pure functions if there are dependencies on operations outside the control of the actual application.
Examples of this would be:
User permissions of the user starting the application
Ports already being used by other applications on machine
Hard drive being full
Files not being able to be written to due to file in use by other applications
Network issues
These are just a few that spring to mind.
I would like to believe that writing code which could never be pure should be grouped and written in a way to maximize other parts of code to fulfill pure function status. So the question is: Is it possible for an application to be built by only pure functions and if not what are things that can stop this and how would we approach these matters?

Best portable method to implement an algorithm on Website

I came across this problem,I have got an algorithm that I need to implement on websites.
The server side scripts may differ and it can be PHP,ASP.NET etc.
All I need to do to is to deliver the binary(I need it as I dont want this security algorithm to be open and viewable) that can comply by every type.As per me solution can be (Please correct me if I am wrong):-
Implementing binary according to OS.
Implementing Algo as per every type of script.(Tough and less portable)
Please suggest if there is other way round or please close this question and redirect me to any earlier question asked for this situation.I am new for this.
Thanks
I would suggest you to use Java Server-side technologies to implement your algorithm. You can write your algorithm as a java class, which can be called from a Servlet or a JSP or even any other technologies over http protocol. The main reason why I suggest using java is:
1) it is platform independent, so your 1st point:
Implementing binary according to OS.
You dont have to worry which OS the client would use.and it can be ported to other OS very easily.
and
2) it will be very secure, once you compile, a class file will be generated, which can be delivered. it cant be opened and viewable.

Rails or node.js for Restful API

I have an existing website written in ASP.net, I have recently switched to Mac full time (With Windows in Boot Camp), and need to write a public API for accessing my website's MySQL data. I primarily want to use the API to allow building an iOS application.
I am interested in learning either Ruby on Rails or Node.js, I haven't used either of them yet.
Which language would be better for me to learn?
Rails is a relatively mature web framework based in Ruby and is designed for handling object-mapped data persistence in a relational database backend.
Node.js is much newer on the scene, and unlike Rails, is a more bare-bones package that allows for server-side Javascript applications thanks to a pretty tight HTTP(S) API. Node applications are by nature event-driven, which may or may not be ideal for your application.
Since it seems that you'll need data-persistence (you mentioned accessing MySQL data...), Rails might be easier to get started with, as it comes packaged with all the things you need in this respect and is designed to facilitate this sort of application.
If you you don't really need relational data persistence, Node is probably a better bet as it stays out of your way and lets you decide how to handle things. It's important to note that Node is a much more bare-bones "framework" than Rails -- if you want something slightly higher-level but still lighter than Rails that runs on Node, express is good place to start.
Still, if you want to try Node (I will confess: it's a lot of fun!), it's totally possible to access MySQL in a nice, event-driven (non-blocking) way. Here are two modules that will be helpful:
node-dbslayer
node-mysql
Neither is a language; they're frameworks. There is no "better", there may be a "more suitable" (probably not in this case).
Which would you prefer to develop in, Ruby (Rails), or JavaScript (node.js)?
Would you like to transfer that knowledge to a different job without rampup time (Rails)?
Would you like to learn something a bit more esoteric, event-driven (node.js)? (Ruby has Event Machine, but IMO node.js takes it a bit further.)
So this new application is just a middle man between your ASP.NET thing and your other clients.
In that case, totally use node. Node is great at being networking glue. Node scales great with IO bound applications (i.e. being a network middleman).
If you're going to use node then you probably want to look at express to make it easier.

GUI Development - Free Tools

We are looking at building a GUI application having the following attributes -
* approx 100 screens
* approx 200 users
* interfaces with 8 different legacy applications (protocols tbd but we are assuming web services will be exposed)
* has a separate data store for storing authentication and authorisation info along with few other information.
We are trying to avoid spending any money on procurement of physical servers, application servers etc.So we thought instead of building a web application , we could develop a standalone GUI based application that could be deployed on users' desktops considering that the users are less and the access to this application is confined to our organisation.
We were thinking of Java Swing as one of the options to consider. Is it a good choice? Please advice on the other options that we need to think about. We thought MS products would involving licensing costs to buy the product hence we are looking at developing the application using some free software tools.
Thanks a lot!
Regards,
V
I really like Swing and have written many applications with it, but I'm not sure the argument that you will get a cheaper solution because you have to procure less physical servers is a good reason for it.
There are other ways to use computational capacity on the client side. For example you could use SmartGWT. With a browser based solution you get easy deployment and have no problems with client-server communications (i.e. no proxy hell). It's possible to get all of this solved with a standalone GUI, but it requires experience and time (== money).
I think a standalone GUI is a good idea if you need a "rich" user experience and tight integration with the native environment. Web apps are generally less snappy and not so well integrated. If it's just about editing data on a server, the standalone GUI does not really have any advantages.
I won't rehash what we talked about in the comments. It seems you are focused on java gui solutions. With that said there are alternatives for developing gui applications with free tools. Just to name a couple, you could develop your client with Adobe Flex or Adobe Air, use python, php or ruby and GTK, or develop a windows GUI app using c# if your target os is windows. Similar to Air, Microsoft Silverlight is an option. Java is not the only solution.
Many Java gui apps are written using swing. The major competitor to Swing has long been SWT which is used for eclipse and by IBM. Javafx is a newer/alternative platform Sun/Oracle is behind, you should also look at. To be complete, you should also take a look at qt jambi that lets you develop in java against the Qt framework.

Client-side image processing

We're building a web-based application that requires heavy image processing. We'd like this processing load to be on the client as much as possible and we'd like to support as much platforms (even mobiles) as much as possible.
Yeah, I know, wishful thinking
Here's the info:
Image processing is rasterization from some data. Think like creating a PNG image from a PDF file.
We don't have a lot of server power. So client-side processing is a bit of a must.
So, we're considering:
Flash - most widespread, but from what i read has lackluster development tools. (and no iPhone/iPad support for now).
Silverlight - allows us to use .NET CLR, so a big ++ (a lot of code is in .NET). But is not supported for most mobiles ( rumored android support in the future)
HTML5 + Javascript - probably the most "portable" option. The problem is having to rewrite all that image processing code in Javascript.
Any thoughts or architectures that might help?
Clarification: I don't need further ideas on what libraries are available for Silverlight and Javascript. My dilemma is
choosing Silverlight means no support for most mobiles
choosing Flash means we have to redevelop most of our code AND no iPhone/iPad support
HTML5 + Javascript we have to redevelop most of our code and not fully supported yet in all browsers
choosing two (Silverlight + Flash) will be too costly
Any out-of-the-box or bright ideas / alternatives I might be missing?
This is the sort of issue that software architects run up against all the time. As per usual, there is no ideal solution. You need to select which compromise is most acceptable to your business.
To summarise your problem, most of your image processing software is written in .NET. You'd like to run it client-side on mobile devices, but there is limited .NET penetration on mobiles. The alternatives with higher penetration (eg. Flash) would require you to re-write your code, which you can't afford to do. In addition, these alternatives are not supported on the iPhone/iPad.
What you ideally want is a way to run all your .NET code on most existing platforms, including iPhone/iPad. I can say with some confidence that no such solution currently exists - there is no "silver bullet" answer that you have overlooked.
So what will you need to compromise on? It seems to me that even if you redevelop in flash, you are still going to miss out on a major market (iPhone). And redeveloping software is extremely costly anyway.
Here is the best solution to your problem - you need to compromise on your "client side execution" constraint. If you execute server side, you get to keep your existing code, and also get to deploy to just about every mobile client, including the iPhone.
You said your server power is limited, but server processing power is cheap when compared to software development costs. Indeed, it is not all that expensive to outsource your server component and just pay for what you use. It's most likely that your application will only have low penetration to start off with. As the business grows, you will be able to afford to upgrade your server capacity.
I believe this is the best solution to your problem.
Host you image processing on Amazon E2C, Azure, or Google. IIRC E2C has many common image processing problems packaged and all ready to go.
Azure probably more familiar ground in term of sharing code as a web service
You just pay for CPU cycles and transfers/storage etc
I'm sure there will be Silverlight and JS people posting examples. Here are some image editors written in actionscript:
Phoenix
PhotoshopExpress
There is an ImageProcessing library to start with.
Plus PixelBender is available in Flash Player 10, it's fast, it runs in a separate thread
and people do some pretty mad things with it.
HTH
Some help for the Silverlight part:
There is an Silverlight image editor called Thumba.
And Nokola recently made one called EasyPainter and he will also provide the source code in the furure.
For the image conversion I would recommend the open source library ImageTools that also includes some basic effects.
Silverlight has a class for pixel manipulation of bitmaps called WriteableBitmap. The open source library WriteableBitmapEx is a collection of extension methods for Silverlight's WriteableBitmap. The WriteableBitmap API is very minimalistic and there's only the raw Pixels array for such operations. The WriteableBitmapEx library tries to compensate that with extensions methods that are easy to use like built in methods.
Pixel Shaders can also be used to make some fast and advanced effects. Although they are limited by Shader Model 2 shaders can be used for fast bluring, tinting and such things.
DISCLAIMER: I consider myself as an advocate of the Flash platform. I admire Silverlights huge potential as a technology to deploy almost any .NET content through the browser, but it has low penetration, is horribly marketed and -although perceived as such by many (mostly people who don't know either Flash or Silverlight)- is no competitor of Flash, as much as Flash is no competitor of Sliverlight. The idealist in me loves the idea of doing everything in HTML+JS using a standard, instead of relying on 3rd party proprietary software. But the truth is, JS is slow and the API is limited, and implementations of JS, HTML and CSS are terribly inconsistent accross browsers.
If you really wanna stick to .NET and are so interested in targeting the iPhone and its siblings, then you might wanna check out MonoTouch.
Still, even though this may surprise you, I am going to tell you to use Flash. :)
Why? The image processing bit is the smallest part of your application. Whatever it is you are writing, I am very sure of that. I don't know about Silverlight, but in Flash the filters used by "Thumba" and "EasyPainter" can be created within a day, most of them simply using ConvolutionFilter, ColorMatrixFilter, DisplacementMapFilter and BitmapData::paletteMap or even simply by applying one of the other filters Flash offers out of the box. Any additional things can be created using PixelBender, which was pointed out by George. The kernel language is a subset of C, so porting classic filters shouldn't be too time consuming. Also alchemy (an LLVM backend targeting Flash Player 10) would be an option worth investigating, although it's not very stable yet.
The biggest part of your app will be a lot of GUI design, GUI implementation, Business Logics etc. Flash is really great when it comes to simple, yet reasonably fast image manipulation and with the Flex framework and MXML you have a powerful tool to productively create the GUI of your app, that can interoperate very well with a multitude of server solutions for virtually any platform.
Also, Flash has a great and active community, offering tons of tutorials, code snippets, libraries and frameworks, and a big ecosystem, with cross-compilation tools to deliver flash content to other platforms (including the upcoming Flash CS5, or the mentioned Elips). I don't understand, where you got the impression, that the Flash platform lacks developement tools. The difference to the .NET suite is that they are provided by a multitude of vendors. The upcoming Flash Player 10.1 was already pointed out by George, but never the less, I wanted to stress, that this makes many of the cross-plattform considerations obsolete.
Last but not least, I'd like to point out Haxe. It allows compiling to SWF, but also to C++, using the very same API provided by NME, to target the iPhone. Also there's work in progress on an android backend. If you're aren't playing to launch within the next 4-5 months, then this is definitely an option.
Your issue is a perfect target for the Haxe programming language. Haxe is written for the web and can compile to JavaScript, Flash and Objective-C (possibly Java/.NET soon).
So you do not choose which platform you are going to invest in but in which language. Haxe is easily adoptable for an AcitonScript programmer.
It makes no sense to run your imageprocessing algorithms in a JavaScript sandbox when Flash is available because it will be much faster. It makes also no sense to run heavy image processing algorithms on a mobile device like the iPhone with JavaScript. I would only support JavaScript as the worst fallback solution.
If you do not like to use Haxe I would go with Flash. You can deploy your Flash application for the iPhone aswell if that is your problem. This is also very great because you get native ARM code. There are actually great tools for professional Flash development available. FDT and IntelliJ IDEA are two of them. The best Haxe IDE is probably FlashDevelop at the moment of writing.
So I would definitly not use JavaScript as the only solution. Haxe is perfect for what you try to achieve. If you do not trust or do not want to invest in Haxe you can use Flash because of the iPhone/iPad export.
Depending on your usecase I would also encourage you to look at cloud hosting like Amazon EC2 and Google AppEngine for instance. Hosting costs are cheap and scaling will be easy for your task. The experience will be much better when it comes to complex operations that can take even a lot of time on a desktop system.
In addition to other answers, another option may be a hybrid solution. For example, use Flash/Silverlight for the majority of your target audience and use server-side processing for those that don't support it (or you could create a native app for iP[hone|ad])
You may have to do something like this anyway as the mobiles you are targetting may have insufficient processing power depending how complex your image processing gets.
Of course you still have the option of upgrading your server which, although you've currently discounted, is probably far cheaper than spending development time creating/deploying/testing a client-side solution.
You can use Silverlight for all Silverlight enabled clients and for non Silverlight clients, do the image processing server side. Since the Silverlight code is C#, you can double compile it to make (mostly)the same code work as Silverlight and non-Silverlight (i.e. server). This gets you the best of both worlds.
You don't say what language "all that code" you'd have to rewrite is in. Might a semiautomated translation to Javascript be practical?
Perhaps you could start out server-side, as CraigS suggests, and then move functions into the client over time instead of rewriting all at once.
Have you checked the editor of Pixlr.com ?
Take a look at their API as well..
The best solution is to use silverlight (so you already have the code ready). If the client can't run it (mobile phones, etc) then process it server-side.
It's the best compromise.
Depends on the type of image processing and the end user experience you are targeting.
As you are looking to target mobile phones your image processing will need to take into consideration the type of handset the user or the receipient has (if messaging via SMS/MMS), as different handsets have different resolution screens and handle different image formats for main images and thumbnails.
I'd suggest that you consider a hybrid cloud architecture as was mentioned in the Microsoft PDC keynotes this year. This would enable you to have your own server(s) to support your application, but if you require additional capacity due you scale out into the cloud using AppFabric.
Additionally, to maximise the market availability of your product pulling the image processing to a common reusable infrastructure allows you to target different platforms, exploiting the positives in each.
I have worked on a solution that hosted its image processing and delivery infrastructure server side and then built different UI offerings allowing sales via desktops, MNOs and AppStores. It can work and from a business perspective can offer economies of scale benefits.
Why not mention Java Applet ?
Good sides are:
almost all browser support ?
need install JRE ?
all OS support
Java provide Java Advanced Image kits, but if c++ dll can be called, that is best (JNI can call c++ dll )
In Python, one of the most popular libraries for image processing is pillow. Through the pyodide project (python running inside browser via emscripten), it's possible to use libraries like pillow and numpy for image (or matrix) processing, and convert the output to a base64 string (via Python standard library). This can then be passed to your <img> html element, either native JS document or with a library like React.
The way I see it, there's no one solution that meets all of your needs. Your best option, imo, is to go with Flash and hope that Adobe sets an agreement with Apple to get Flash on the iPhone/iPad. The major downside, of course, is you'll have to rewrite much of your code.
If the mobile sector isn't absolutely critical, then choose the Silverlight option for reasons you mentioned already. You could also use Silverlight in an out-of-browser mode to work as a desktop application.