Cloud Computing Conundrum: An Interview with Chris Moyer
Like this article? We recommend
Like this article? We recommend
Like this article? We recommend
There is a sense in which cloud computing is nothing new. After all, at the end customer level, all Internet services, from the World Wide Web to email, even file and print sharing, involve clicking a button and having computers "somewhere over there" do something.
Then again, these were typically one server that ran all the time. Sure, there were a few supercomputers in New Mexico simulating chemical explosions for the Department of Energy, but it wasn't until the dot-com era that we really saw a server farm. Google's architecture of redundant, independent servers was certainly a breakthrough—but even that meant physical servers on physical devices.
Then in March 2006, Amazon unveiled EC2, the "Elastic Cloud Computing Platform." All of a sudden, anyone with a credit card could create and direct virtual servers at will.
If only they knew how.
Enter Chris Moyer, a graduate from the Rochester Institute of Technology, who had his first internship building massive scale applications for the school using traditional virtual servers running VMWare. Shortly after Amazon announced EC2, he went on to work with Mitch Garnaat, the creator of boto, the python library for communicating with Amazon Web Services, or AWS. Chris spent the next five years directly developing, deploying, and consulting on virtual applications for businesses of all sizes, from small startups to Fortune 500 companies.
Currently the director of technology for Newstex, LLC, Chris specializes in developing applications for the cloud and helping others do the same. Chris recently sat down with Matt Heusser to answer some quetions about developing for the Cloud.
Matt Heusser: It's hard to imagine a high school or college student saying, "When I grow up, I want to virtualize servers!"
Chris Moyer: They don't?
Matt: But I imagine something must have happened to excite you about this new field. Can I ask: What turned you on to cloud computing? What made you want to pursue it with such vigor?
Chris: I always wanted to make something that people used in their everyday lives, something that I would be able to point to that everyone used on a regular basis and say, "Hey, I wrote that!" That really meant I had two options: either make a desktop application, or go under the hood and work on the applications that really powered everything. I was bored with traditional environments, and when Amazon started offering up these new "virtual servers as a service," it was new and hadn't been fully explored and developed to death, so of course I wanted to take a crack at it. And after a few months, I realized everything I did when I first started working in the cloud was completely wrong. That's really why I decided to write a book. I don't want everyone else to have to go through six months of doing things wrong before they get something usable.
Matt: I can see where your passion comes from. Take us back to 2006 when Amazon S3 actually went live. How were you introduced to it, and what did you do first?
Chris: At the time I was working at RIT, building some high-scale applications using traditional data centers. Then my boss's boss came to me saying a friend of his needed some help with a small startup company he was working on. I've been trying to keep up with all the changes and services they've been releasing since then.
Matt: Tell us about some of the services you've written for the cloud computing platform. What do they do? Why not "just" write them on a more traditional infrastructure?
Chris: Probably the best example is a system I designed that pulls down videos from RSS feeds and processes them, picking which formats to save, doing some encoding if necessary, and even running it through a Bayesian classification system to make sure it's not just spam.
Matt: Can you tell us a little more? Who is the customer for the service? Again, why not "just" use a real server?
Chris: The service is actually used in a number of locations, most notably being the mobile applications developed by my current employer. We didn't want to use just normal hardware because that would limit our expandability when our applications take off. When you go from 50 videos a day to converting 50,000 videos a day overnight, you want that scaling to just happen for you. With AWS, that's completely possible, as long as you build your system the right way.
Matt: That brings me to a larger question: how can we make the decision to go cloud or not? What factors should we be considering?
Chris: Anything can be moved to the cloud—it just depends on how much time you have to work on it. Obviously most people don't have an infinite amount of time to move their system to a cloud-based environment, so you just have to weigh the pros and cons. Long-term, it will be cheaper and more scalable, but it may take you an extra three months to develop.
Matt: Let's say that I'm a little bit of a more…seasoned developer. My manager walks by an casually mentions that the next version of our server-based application needs to "automatically scale up" with use and that we'd like to "not have to worry about hosting," so he tells you, "We're going to use cloud computing." Now what?
Chris: Just like with anything, you have to do some planning and research. You really have to know what you need out of your application before you can even know if it's a good fit for cloud computing. It's important to plan, but it's also important to just try some things out and see what you can do. Amazon's cheap services mean you can spend a dollar and learn a lot about virtualization.
Matt: Well, let's say you want to experiment with EC2 to create a scalable webserver. Where do you start? Specifically, what should I do? I'm looking for webpages and specific advice. Say I want a Linux webserver and a replicated MySQL backend. Now what?
Chris: Well, I can't give away all my secrets or you wouldn't buy my book now would you (chuckle)? But seriously, the first step is always research—for example, if you're looking for how to just get started with a simple micro instance, Apache, and MySQL, you can head over to my blog. Starting from there, you can build your system out to be more flexible and scalable as you need.
Matt:I know I just asked you how to create the infrastructure for Facebook; perhaps that's a bit ambitious for a short interview (laughs). Along those same lines, when you started creating web services, what would you say were you first few mistakes? To put it differently, what common mistakes could you help our readers avoid?
Chris: Well, the first thing to avoid is making assumptions. For example, you said MySQL. Do you really need MySQL or do you just need a database? Does it need to be relational? Do you really need to store files on a filesystem, or will an API work better for you? Once you break down and know what you actually really need for your application, you'll know more what patterns you can use, and then you can figure out what cloud services are best suited for those patterns. I find drawing systems out before developing helps a lot to find the single points of failure, and then you can try to eliminate those before you start to develop.
Matt: Let's go back to that “spend a dollar and learn” idea. Do you have any specific exercises you suggest (or can recommend) about learning how to use a particular cloud service? Something with a trail of breadcrumbs that a programmer can walk through, and when he's done—shazam—web service?
Chris: The boto website has a lot of tutorials for almost all the common services provided by Amazon. And there are more included with the source code for boto itself. For anything more in-depth than that, there are a few good tutorials in the final section of my book, which guides you through how to set up a quick blog using AWS with two different methods.
Matt: Speaking of the Shazam web service, can you tell us about how it feels to set up a scalable server? I mean, I imagine you've got have some sort of load-balancer sitting out front, taking connection requests and routing them, maybe measuring the amount of load and spooling up a new server if needed. That sounds like a lot of code to me. Help us understand. At the code level, what does that look like?
Chris: I'm sure you've seen the movie “Independence Day” when the crazy guy sits on the nuke saying, "I just wanted to feel the power between my legs." Well it's kinda like that: You realize there's almost no limit to what you can do; it's a lot of power for one DevOp. It's not a tremendously hard thing to do either, because most of the key components you need are already provided for you by cloud providers. Using my book, you can quickly go through and find out exactly what solutions you need for what problems.
Matt: Your book has a substantial focus on migrating existing applications to the cloud. I was struck by this sentence:
"Another large misconception is that if you use a cloud provider, your system will be highly reliable and always available. This is not true because most cloud providers use nothing more then commodity hardware for their underlying hardware. Although this hardware is no more or less likely to fail than your personal desktop, it’s also under a lot more load, and it’s used by more people."
In other words, a fancy app moved to the cloud but not re-architected might be slower and more likely to fail. So you go on to recommend re-architecting an app. If I've got a database running Oracle, how do I even do that—or perhaps I should ask, what's the first step?
Chris: Actually, Amazon and Oracle have teamed up now, but that's really not the point. The main thing to do is break down your application into its base components. You have to throw away your traditional misconceptions and open up to new ideas. Before you decide how you move your application, you need to know what your application really needs to do. The next big step is to diagram out your system and determine which pieces will work and which will have to be re-designed. The biggest thing to remember is that even if you do toss up your application almost directly onto the cloud, you don't want to do it all as one piece.
Say, for example, you have a complex application that uses a database and stores large files on a disk. You'd want to split that out, host the database on a separate machine—or even multiple machines—and put those "large files" into something like S3, not just the local filesystem. By starting there, you've already increased your reliability and expandability. Your end-goal really is to make it so you can launch more servers that act together so you can scale out instead of just scale up. That is to say, add more machines to the mix, not just one bigger machine. The idea is to not store anything long-term on a physical machine, but instead to store it someplace else, like an external database, or file-storage system like S3.
Matt: Speaking of proprietary databases, many companies I know with proprietary data are hanging on to their data centers because the cloud is scary and public. Is that a valid concern? If it is, what kind of applications are a “good fit” for these types of services?
Chris: Well, if it was, I don't think Amazon would have been approved for government uses. We're talking about the folks who are so paranoid they wrote their own OS now saying it's ok to use AWS. Billions of people use Google, and they admit they use your private info. Amazon, on the other hand, goes through massive security protocols to make sure your data is secure.
Matt: Thank you for participating. Where can we go for more?
Chris: Obviously, you can check out my book. There is also a fair amount of free materials and examples on my blog at http://blog.coredumped.org.