Slashdot Mirror
Ask Slashdot: Building a Web App Scalable To Hundreds of Thousand of Users?
AleX122 writes "I have an idea for a web app. Things I know: I am not the first person with a brilliant idea. Many others 'inventors' failed and it may happen to me, but without trying the outcome will always be failure. That said, the project will be huge if successful. However, I currently do not have money needed to hire developers. I have pretty solid experience in Java, GWT, HTML, Hibernate/Eclipselink, SQL/PLSQL/Oracle. The downside is project nature. All applications I've developed to date were hosted on single server or in small cluster (2 tomcats with fail-over). The application, if I succeed, will have to serve thousands of users simultaneously. The userbase will come from all over the world. (Consider infrastructure requirements similar to a social network.) My questions: What technologies should I use now to ensure easy scaling for a future traffic increase? I need distributed processing and data storage. I would like to stick to open standards, so Google App Engine or a similar proprietary cloud solution isn't acceptable. Since I do not have the resources to hire a team of developers and I will be the first coder, it would be nice if technology used is Java related. However, when you have a hammer, everything looks like a nail, so I am open to technologies unrelated to Java."
http://www.codinghorror.com/blog/2010/01/cultivate-teams-not-ideas.html
http://evergreen-ils.org/opensrf.php I do not know much about it. Here is Dan Scott's first paragraph form his 'Easing Gently into OpenSrf' article: OpenSRF is a message routing network that offers scalability and failover support for individual services and entire servers with minimal development and deployment overhead. You can use OpenSRF to build loosely-coupled applications that can be deployed on a single server or on clusters of geographically distributed servers using the same code and minimal configuration changes.
Just use Heroku. Honestly you DO NOT need to worry about this problem. If you don't make enough money by the time you get 10,000 users to hire someone to solve this problem for you then your idea is not as great as you think it is.
Before going all-out to reinvent the wheel on yet-another-next-big-thing web app, why not roll out a proof-of-principle version letting someone else competent do the "heavy lifting" back-end work. Use an existing cloud/hosting service like Amazon EC2 (they'll do a lot better on the basic back-end stuff than your "I'm incompetent but building a cloud app anyway" approach). After you get your first hundred thousand users, and have investment rolling in by the gazillions, then you hire your own crack team of cloud experts to design your own custom back-end solution (or just sell out for a couple hundred million to whatever group of suckers thinks your zero-dollar-per-user profit model will start paying off once they hit the million-user mark).
I would likely build a front-end using a couple HAProxy load balancers hitting an Apache cluster running opencluster. Use red-black trees with mySQL and cluster a few databases across multiple locations. I would build the front-end with Python and html5, as well as using iphython for cluster controls and other fun stuff.
In my case I have a rack of HP p-class blade servers that use an Amazon EC2 Centos box to route inside/outside of EC2. When we test something out we use my cluster at home, then when we roll an app or website out we keep it at my house. If the load gets high, then we simply modify the cluster to bring up slave web servers, cache servers, etc. In our case we build the backend first and can roll out an app or web service for very little money or resources, but if we have success with something we just leave it on EC2 since it can likely pay its own bills.
When the only tool you have is a hammer, every problem looks like a nail
Youtube was a lame app with basic mysql setup. Same with Facebook. When it took off, they hired gold people and fixed the scalability issues. Twitter didn't exactly put scalability first either.
So get real. Don't worry about "hundred of thousands of users", but about getting something decent out there for users to try. If users come, you'll get scalablity sorted out.
Do not plan for hundreds of millions of concurrent users at once right off the bat. That's the very common error a lot of startups make. You do not have such a large userbase. It will take some time until you have.
Think smaller and scale up when your idea takes off. Set yourself concurrent user milestones when you rethink your architecture. You will also have to rethink the iron your stuff runs on and that may dictate what kind of technology you use when you reached your hundreds of millions goal.
Technology is interchangeable. It's a tool and you choose the best tool for the job and at the moment you have no users and might as well start off with the usual suspects. JSP/Struts, JSF, whatever you are most comfortable with. If in the long run you do find that this is not sustainable and you need to shift to another technology then you can hopefully afford to hire people who know it.
You really, really should set yourself userbase milestones, plan ahead for reaching them and be prepared when you reach them. For that you need a lot of information. Log how much time users spend on what functionality you offer because this also has an impact on your UI design when you go big. It also has impact on what technology(-ies) you use.
I usually bill big when I give advice such as this and help setting up a plan when to do what. Your problem is less one of technology but a business one. Think like a businessman first and like a techie second.
20 minutes into the future
Java... ok, why not. I would take a look at Cassandra and Zookeeper to get the ball rolling. You'll need a good load balancer; nginx or haproxy since I don't know of a good one in Java. I assume a bunch of tomcat servers for the actual app. I suppose jboss messaging to keep with the java theme.
You can get all that on one machine for development, then for deployment you can flexibly adjust the number of db servers, queue servers, load balancers and app servers based on anticipated load. If you're extra-cool you can deloy to a cloud and dynamically allocate servers as-needed.
Been there, done that. Got the t-shirt. It's fun. Enjoy it.
Spend an extra day or two thinking about exactly how you're going to handle logging. It will be worth it.
Can't do it yourself, then get partners. Set up an equity agreement.
As far as tech this is no longer new territory. Create server images for a cloud host such as AWS or Rackspace. Bring them up or down with Chef. Concerned about Database? Figure out if you really need a relational database. If not look at a high performance NoQL DB or something that is more or less always in Memory (such as Mongo).
This sounds very much like premature optimization. You may end up designing a very scalable application and have the project fail due to too few users. If the actual number of users turn out to be an order of magnitude less than what you can handle on a single host, then all that scalability work was wasted. I think you have better chance of success with a quick proof of concept, which isn't very scalable.
It is ok to think about scalability before you have the users. But don't waste time implementing the scalable solution for a non-existing user-base.
Do you care about the security of your wireless mouse?
Probably the worst thing you can do is start with some complex clustered architectural design.
Just start on a single server with technologies that are scalable, and design with future scalability in mind. Also design in the ability to capture detailed performance metrics of every tier. When, and if your application usage grows, scale the parts of it that need scaling.
The biggest issue with scaling is usually the database, and for applications where you are just using the database as a simple persistence store for user settings and simple small data sets, you are probably best to go with one of the many scalable "NoSQL" type solutions such as MongoDB, as they've got scalability baked in for free. If you're trying to run heavy duty analytics that join and aggregate massive datasets, there are single DB clustering solutions, but they aren't cheap. You can always scale out SQL databases horizontally, but then you've got issues cloning and replicating, though there are a lot of products in that space, both free and commercial. A cheap place to start would be with PostgreSQL, which appears to have multiple open source replication products.
I don't think there is anything inherently limiting to sticking with Java. It's what you know, and the toolsets are deep and rich. No, it's not the hot new thing, but sometimes that can be a good thing.
Facebook did it on PHP. I sure wouldn't have used that, but it shows you can do more with basic technologies than you would expect.
The Java environment was built for that kind of thing, Spring, Hybernate, etc, so if you build in that, you can be reasonably sure your system will be scaleable.
Keeping session state in RAM will make your life harder.
Even with a 'slow' technology, you can always add more servers. The difficult bottleneck is the database, and that can be an intractable problem depending what your goal is.
"First they came for the slanderers and i said nothing."
If you're aiming for as many users as you say, then it'll take awhile to get there and you'll have plenty of time to hire folks along the way. At that point, you can go ahead and worry about re-architecting everything. First things first though, especially if you're by yourself: get it up and running with whatever technologies you do know. Once it starts to take off, you can hire people to rewrite it and redesign it around best practices.
It's not the simplest path, but without bringing in outside investors who'll have the capital to allow you to hire the team it sounds like you need, I don't see what choice you have.
While there are a number of good tools out there for working with scalability, more important than any particular tool is building your application in such a manner that it's easily parallelizable. In a Web app, a core principle to keep in mind is that the more stateful the application server-side, the more difficult it is to scale, and so designing your application tiers in such a way as to decouple requests is key. Limit the amount of session state the server has to keep track of, and you'll be able to load-balance request handling smoothly.
I salute you for your ambition and determination. I hope you get to realize your vision.
Now, as I read your question, I remembered an interview I saw a few days ago with Ben Kamens, one of the engineers working at Khan Academy, talking about scalability and things like how they manage their operation and the spikes of growth they have experienced in the past. It's a little light in technical details, but you may find it interesting: Root Access: How to Scale your Startup to Millions of Users.
One thing I'd like to mention is that when you hear someone else talk about the things they've done and how they have done it, it's easy to see it as an advertisement for a particular technology platform (AppEngine and other Google machinery in the previous video, for example), but that's not the thing to focus on. Whatever choices other people have made, the good thing is that their advice can be useful no matter what choices you end up taking. I know this seems like such a trivial thing to say, but evidence suggests that a number of people miss this basic concept, and then discussions quickly degenerate into pointless noise about concrete technologies, instead of the ideas.
I'd also recommend that you pay a visit to Google Developers youtube channel and type something like "scale" or "scalability" in the little channel search box. You might learn a few things from some really smart people who have confronted very real situations regarding scalability.
Best of luck to you, my friend.
I suggest you look at the CQRS pattern. A good Java implementation is http://www.axonframework.org/. The advantage is the CQRS pattern that it is fairly simple, but highly scalable. So you can start small and simple with the confidence that you can tweak and optimise in the future to scale as required. There are good tutorials and support too. My team is using it for an industrial application and we have found that it has been very robust. It might take a bit of work to get your head around the concepts, but it is worthwhile in the end.
Great advice. I see they helped you remain humble also
Many sites with very very large userbases use Java extensively in their stack. Including eBay, PayPal, Amazon, Tumblr, LinkedIn and Google.
Millions of page views a day is a small to medium e-commerce site. I was doing a million with Perl back in 2002 on a two CPU 1U machine.
Tumblr gets something close to a billion, as does anyone in the top 100.
This. I have found it's best to avoid phrases like "why not shut the fuck up," "why not eat shit and die," and "why not stick your unsolicited advice up your ass." People do not react to these phrases as positive suggestions as intended, and they immediately go on the defensive. Instead, try sarcasm.
keep it simple stupid
the more complex you make the app, the bigger the load on your infrastructure and bandwidth
if you follow google's lead, they developed everything in house. same with pixar, which develops software to handle very high end graphics performance, and even linux started off by taking a problem and solving it with a home grown solution
if you want a specialized application to handle that many users without running into software performance issues (nevermind server infrastructure and bandwidth, which can probably be gradually improved), you want to make it efficient... so you will probably need to develop it yourself
if you use off the shelf packages like wordpress and the like, they are full of all sorts of features that you might not need but will still pay for performance-wise
many people will try to tell you that there is no point reinventing the wheel and that existing wheels will always be better than anything you can come up with, but they are full of shit. if everyone stuck with that ideal we would all have wooden wheels on our cars. there is a lot of merit in reinventing wheels, not only to make better wheels, but in understanding wheels to learn how to better use them. be a little selective about where you want to start customizing from... i wouldn't recommend reinventing the operating system, although google did (based on the linux kernel) and they are reaping the rewards of a more efficient search platform than might otherwise have been possible.
if you're handy with microcontroller programming you might be able to make a pretty efficient microcontroller-based server cluster, sort of similar to what HP is doing with their new SOC blade technology. microcontrollers and SOC are the future, so if you want to get involved in future tech today, pay attention to what is going on with ucs... a simple example is sheevaplugs and its derivatives. this is also where linux probably has a major leg up on windows because microsoft has been so focused on the x86 platform that (even with the recent release of WIndows RT) they are lagging a ways behind linux in multi-architecture support (have to wonder how much of the linux kernel has been plagiarized in WinRT).
other things that affect scalability and performance include the efficiency of algorithms... if you haven't done a CS degree, go onto youtube and watch lectures on data structures and algorithm optimization. there are free CS lecture series from MIT and UNSW that I know of. Richard Buckland of UNSW also makes the lectures a little less boring with his antics.
how you develop your app will also depend on your goal to get 100,000+ users on the site...
security is probably the hardest and most significant hurdle you'll face... if you fuck security up (either the app isn't secure enough or it's a pain in the ass for users to authenticate) then your app will be a flop
you also need to think like a user, not like a developer... this is probably where having a small team will help at some point (a few eyes with different perspectives)
many developers fall into the trap of developing software that is easy for the programmer and thinking that the user will get used to it... which is fine if you have a monopoly. unfortunately by the time you have 10,000 users, your idea will be copied to create competition, and if they do a better job with the user experience you're dead in the water.
make sure you are standards compliant. use the HTML 5 and CSS 3 validators, but i would recommend avoiding features that aren't also in HTML 4.01 and CSS 2.1 until HTML 5 and CSS 3 become fully implemented and debugged. the exception would be that if you want a feature that would otherwise require flash or java, use html5 instead of flash. if you want 100,000+ users, don't use flash or java!
i would use a linux distro such as debian with all the fat trimmed. it should be obvious, but don't use a WISA stack.
keep your service clear of advertising, 3rd party cookies and any 1x1 hidden iframes. don
and vector graphics because they scale like a leprotic fish.
Regardless of which language or platform you use, a common bottleneck for web applications is the database resource. Most developers don't take large scalability into consideration when building the service architecture. If you plan to scale large in the future, I recommend you stop thinking of the database as the main source for all queries in your system. The basic idea is that costly and complex queries/searches can be given to an external scalable service. Take for instance, the Solr project (http://lucene.apache.org/solr/) which is a third party indexing tool that can be easily integrated with any other platform. You can design your system's database with the basic table relationships with primary keys, foreign keys and the occasional index. Any more complex table relationship, queries and searches can be delegated to this external indexing service. It will index whatever data you give it, in whatever manner you need, and return a list of results for you to easily find primary keys for direct access to your system objects. Think of it as your own personal Google indexing service... Solr is an Apache open source implementation. Once you understand this concept, you can keep you application's internal database very lean and simple, with just enough indexes and primary keys to get instant access to entities.
Julio Henrique Morimoto juliohm@gmail.com
Based on my experience at a fortune 100 company with a heavy interest in Java. Don't use Java. Use PHP or LUA as a cgi. Your sysadmins who have to keep your application up will thank you.
Write your public and private Apis first. Then implement them quick and dirty. Get feedback. Get users. Keep working on the API to make improvements. As you get more traffic hire good people to reimplement those same APIs on a better tech stack. Runs and repeat. You can even mix and match platforms, just use a smart routing proxy like HAProxy to send requests to the appropriate places. Static files go to a CDN, logins can go to something small but secure, high volume requests can go to a big cluster or IaaS like Amazon or Google for on demand scaling.
API first.
A fool throws a stone into a well and a thousand sages can not remove it.
One company I work for (until I found a sweeter place elsewhere) got rid of their entire dev staff except for the top level designers. An offshore dev team gives guarenteed results, low bugs per line count, and actual contracts to say that. As an added bonus, the parking garage doesn't smell like BC bud anymore.
You might give them, or another offshore place a call. They may be able to get what needs done, with little QA, for pennies on the dollar than it costs to hire people locally.
In short, build something with a best guess of what you will need, pick whatever you like to scale it with, and then when it is successful enough, sell it to another company who already knows how to scale it.
If you have to, or insist on owning the idea and the company behind it, start hiring people to help you. Scalability isn't a one person job after a certain point. In fact, if you want to see if it is really enterprise scalable, don't even just show it to developers, throw the app on a box and let a real sysadmin look at the specs and poke at it. You should be able to glean how scalable and reliable it is by simply measuring the look of horror on his face.
As someone who has written an application that scales to over 1 billion requests per day, let me offer my thoughts.
Scaling your application should be as trivial as launching more application server nodes. If you can't add/remove application nodes painlessly, you've probably done something wrong like keep state on them (this includes sessions).
Don't worry about scaling your application layer at all (within reason). You can always throw more machines at the application side in a pinch, and for a long while it will be cheaper to add servers than to hire someone. When your application servers are costing you more than a salary, hire someone to find the hotspots in the code and make them faster. Until then it's a waste of your time.
Scaling state, aka your datastores, is where the challenge lies. You need to spend a large amount of time sitting down and analysing every operation you plan to do with your data. SQL is great for a lot of things, but you will eventually run into a point where heavy updates make SQL difficult to scale. Mind you, decent hardware (lots of cores, RAM, and SSD) running MySQL should scale to several thousand active users if your queries are not expensive. The Galera patches to MySQL (incorporated into Percona XtraDB Cluster and MariaDB) can give you true high-availability, but you will still have write-throughput limitations.
I would also highly recommend you look into Cassandra (especially 1.2+, with CQL 3), which was built from the ground up to scale thousands of low end machines that often fail (if you can't tolerate hardware failure, you messed up). Cassandra is more limited in the kinds of queries you can execute, more relaxed with data consistency, and more thought is needed ahead of time. On the other hand, it can also be used for global replication, which is something you are interested in. At the very least, having a good understanding of its data and query model will open your mind to the kinds of tradeoffs that must be made to enabling scaling.
Contrary to what others are saying, you are correct to think about scaling now before you even start! Doing a rewrite is costly and expensive in money and time. Why set yourself up for that? Planning for scale before you start is the best time! If you start with a scalable datastore like Cassandra, and structure all your queries to work within its model, it is no more work than doing things in SQL, and you're way ahead of the game!
The most important part is spending time modeling how you will access your data. Think about how you'll avoid hot spots (which make scaling writes difficult), and think about how to make reads fast by reading as little as possible. Think about caching, and how you'll invalidate the cache of a piece of your data without having to invalidate caches for things that didn't change. (Think about updating on data ingestion instead of running statistics later.) If you can't avoid hot spots, make only small reads, and cache independently, you are not done.
Good luck!
Be relentless!
one of those people worried about where to hide all the gold they're going to have someday? "My app will be sooo successful! I need a team of people to make it for me because I don't know how!" how about you make your super successful app, and if anyone ever bothers to use it *then* worry about scaling it up, mmmmk?
that's a terrible idea - the last thing you want is to have a terrible user experience (requests timing out etc) and deal with complete rewrite (which needs to be completed by yesterday) the moment your site becomes vaguely successful.
This can easily kill your project at a critical moment - users don't like to hear "sorry, but we couldn't anticipate that (some blog/the local newspaper/...) would mention our site and people would actually try to use it; please come back in 3-6 months when we have rewritten it for scalability".
Ignoring scalability "until I have users" is a great way to keep costs down while making sure that you cannot ever be successful. If you think that is the financially rational thing to do (because 99.9% of such projects don't succeed anyways) then you shouldn't sink money into a website that is (in your opinion) bound to fail at all. Butr if you are going to invest then you have to invest enough that you actually stand a chance at success (however slim that may be).
If we're doing work for you, how much do we get?
Vote monkeys into Congress. They are cheaper and more trustworthy.
Brilliant idea, that's sure to work.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."