Facebook引领数据中心开源革命

除了你们之外，还有别人在从事数据中心硬件开源的工作吗？

    我们与许多其他项目建立了合作关系，不过我们特别关注的是数据中心的硬件设计。据我所知，目前没有其他项目着力于这个领域。过去的做法是把所有的牌都紧紧捂在胸前，不与任何人分享。不过开源操作系统Linux的成功以及它对市场的影响启迪了我和Facebook的所有人，激励我们去参与这件事。我们希望硬件开源也能产生类似的影响。

你们有没有哪些技术是不愿意“开源”或与别人分享的？

    我们在技术开源的问题上考虑得非常周到。我们分享了数据中心选址的方法。不过我们在分享数据中心模板的时候并没有包含光纤主要接入点的分享，因为我们认为这是一个安全性问题。我们在这种问题上不会选择开源，不过那是因为我们需要保护自己以及我们的终端用户。我们在应用领域的关键创新也不会开源，因为正是那些独特的技术造就了Facebook的独特地位，同时那也是9亿多用户选择Facebook的主要原因。英特尔也是开放计算工程的创始成员之一，同时它也是业内知识产权资源最丰富的公司之一。虽然英特尔的工程师们也（为开放计算工程）做出了大量贡献，但我们并不指望他们会分享他们设计CPU的技术。

你们的设计有一部分实际上简化甚至取消了数据中心的某些功能，很多人可能对此感到诧异。你能解释一下吗？

    可能很多人都会觉得在服务器上贴上一堆塑料Logo是一个好点子。不过有时候想把事情做简单反而很难。人们有时会把事情过度复杂化。一个设计可能看起来非常精美，因为它有各种出色的功能。不过如果后退一步，问问自己怎样能用最少的部件把事情做好，你就会发现，有时简单才是最难的部分。拿有些最成功的移动设备来说，当你把它们从包装盒里拿出来的时候，它们看起来并不像是无所不能的样子。它只是一块屏幕加一个按钮。但是在开机之后，你让它做什么它就能做什么，这时人们才真正体会到这个设计的美和简约。但人们看不到他们为了简化设计所付出的工艺上的努力。

开放计算工程下一步的打算是什么？

    吸引我们的方向有很多，其中大多数集中在数据中心和服务器的设计，而且我们的工作已经扩展到了（服务器）机柜上。存储空间领域有可能升温，而且我们对网络设计也有很浓厚的兴趣。不过在接下来的六个月里，我们的很多活动都会围绕着存储——也就是如何让开源存储真正地改变市场。希望它能帮助企业选择最佳的硬件和软件。未来随着云计算趋势的兴起，大型数据中心运营商的数量会有所减少。我们现在已经到了一个拐点，以后一切可能会变得更标准化一些。不过关于未来最令人兴奋的是，我们现在可以应用业内最好的脑力，来满足计算方面最新、最独特的要求。

译者：朴成奎

Are there any other efforts out there to open source data center hardware?

    We have partnerships with a whole bunch of other projects , but we are specifically focused on the hardware design in the data center, and to my knowledge there are no other projects specifically around this. The old method is to keep all your cards close to your chest without sharing. The biggest project that inspired me and all of us at Facebook to get involved is the open source operating system Linux and the impact it had on the market. We want to have a similar impact on hardware.

Are there technologies that you won't "open source" and share with others?

    We think really, really carefully about what we open source. We've shared how we pick data center sites. But when we open sourced our data center blueprints we didn't include the main point of entry for fiber runs—we felt it was a security issue. So there are some things like that that we don't put out in the open. But that's really because we need to defend ourselves and our end users. The thing we won't open source are the key innovations we have in the application space. Those are the unique things that differentiate Facebook and the reason more than 900 million people come to Facebook. Intel is one of the founding members of the Open Compute Project. It also happens to have one of the richest IP portfolios in the industry. Intel's engineers have made significant contributions [to Open Compute] but we wouldn't expect them to share how they design CPUs.

A lot of people would be surprised to know that some of what you've done with your designs is actually simplifying and taking away capabilities. Can you explain?

    I don't think anyone would argue that putting a bunch of plastic logos in front of a server is a good idea. But sometimes simple is actually really hard. People sometimes overcomplicate things. When you look at a design it might look really elegant because it's got all kinds of whiz-bang features. But when you step back and ask how you can do this with minimum components, sometimes making it simple is really the hard part. Some of the most successful mobile devices don't look like they do anything when you pull them out of the box; it's a flat screen with just one button. But when you turn it on and it does exactly what you ask it to do, then you really understand the beauty and simplicity of the design. You don't see the engineering efforts that went into making it simple.

So what's next for the Open Compute Project?

    We're getting a lot of traction. Most of it is in data center and server design, and we've extended it to [server] racks. The storage space is something you'll see heat up, and there's also a lot of interest in networking. But a lot of the activity in the coming six months is going to be around storage—how open source storage really changes the market. Hopefully it will let companies choose the best of breed from both hardware and software. In the future there will be a smaller number of larger data center operators because of the trend towards cloud computing. We've reached an inflection point where things can get a little more standardized. What's exciting about the future is that we can now apply the brain power to new and unique requirements in computing.