That's very impressive, they're essentially risking enabling their competition by doing this and still they've done it.
This actually will allow the development of 'micro clouds', a thing I wrote about a while back because I don't like the centralization issues of clouds for a number of reasons and use cases.
http://jacquesmattheij.com/node/37
Getting rid of vendor lock-in is also fantastic.
Along with other benefits such as a developer community around their infrastructure and ease of friction around moving cloud hosting providers (i.e. Vendor lock-in) makes this a clear win and an excellent move by Rackspace.
Its basically a tool for managing virtualization clusters, with a ton of nice features: live migrations, redundant storage over DRBD, support for KVM and Xen (I think support for a few other virtuualization/containerization systems is in the works too), scriptable deployments and an HTTP API. It is, as I understand it, used internally at Google (they developed it) for a few things, and judging by the mailing lists there are a number of other good sized deployments around as well.
In a few hours you can set up your own internal "cloud" with Ganeti, and its fairly easy to scale later as the need arises. Bring a new node online, get on the Ganeti master node and tell it the hostname and root password, and Ganeti will SSH in and take care of most of the setup for you.
Swift documentation is at http://swift.openstack.org. Check out the "SAIO - Swift All In One" page to see how to set up a VM to run it. With some very small tweaks, you can also run it on a single slice (mount a loopback device rather than another virtual drive).
Although I don't know as much about it, the compute docs are at http://nova.openstack.org
I'm guessing that both Amazon and Rackspace took a look at MogileFS and its architecture so it would be informative to hear what you liked and what you might have found lacking.
Openstack doesn't match Amazon's APIs. Nova (compute) is used by NASA and will be used by Rackspace Cloud Servers. Swift (storage) is currently used by Rackspace Cloud Files. Swift (my area of expertise) is production-ready code that is scalable to massive levels (think 100-petabyte clusters and 100000 requests per second).
Because benchmarks are very sensitive to deployment details, the only thing I can point you to is comparisons between Rackspace Cloud Files and Amazon S3. Of course, even then there are all kinds of variables like cluster size and network connectivity.
The theory is that I can take a data snapshot of my 20 terabytes, move it to the new provider's network by driving it across town (still the fastest way to move data), "drop it in", switch over DNS and excluding a few lost transactions during switch-over everything should work fine.
The reality is that this removes the pain of proprietary configuration from a data center migration but it's still a painful process to switch cloud providers that involves a lot of work, some risk and down-time. So while this is a great marketing play from Rackspace there is still a level of lock-in when you commit to using a provider and so the risk to Rackspace to go 'open' is minimal but the benefits of being the leader are huge.
If you are correct, this sounds like quite a cynical move to try to hoodwink potential customers into a false sense of security with an apparent promise of platform independence.
A number of us in the open web community have used that term to a great extent to talk about OpenID + OAuth + Portable Contacts + Open Social (http://therealmccrea.com/2008/09/19/joseph-smarr-at-web-20-o...)
Admittedly the initiative has lost momentum (for many reasons and factors) but things are still out there. This just creates confusion and branding collision.
I notice RackSpace is pushing a "TM" on their OpenStack branding, which calls into issue anyone using the term for it's prior and original meaning.
But the wider communities; the customer communities and the developer communities are going to be in that overlap between both infrastructure and application.
And we both know that once your brand goes through 'the wash' of the media it's going to be munged into "Open Stack" as much as "OpenStack".
Sure everything has some kind of name collision but that's just indicates why more deeper investigation is required to ensure such collisions occur between two very different usages. It is disappointing in this instance that didn't occur.
(I'm guessing/assuming you are a RS employee...) you also haven't touched on the concern that RS is asserting a TradeMark on "OpenStack" which brings some concern into the existing publicity for the Open Stack concept.
The problem is that the domains are very close. There's a lot of modern web apps are cloud based and use the open stack (I recall Mike Malone and Joe Stump doing a great talk at PyCon about these topics using that exact term). This makes searching really really annoying.
How about OpenSkies?
Whether it's a big hosting company running a public cloud, a corporate running a private cloud, or a smaller dev/test cloud, they can now all run the same free software stack, and software can move freely between those environments. It's clear this announcement will have ripple effects throughout the entire cloud ecosystem. I hope the project will also boost adoption of cloud-style infrastructures, now that the fear of supplier lock-in is removed.
It was fantastic to see what they've put together and the business model is great. Commoditizing architecture and software really is the next step.
If anyone is interested, I could write up a blog post about my experience, maybe provide a little more insight into the buy-in.
I think that there will be a real sweet spot for smaller organizations who would like an internal system like S3 that offers robust data security with using really cheap disk drives and servers.
Hopefully OpenStack Object Storage will have good support for multi-datacenter replication.
Or should someone like me even care?
Right now, to move to a new host, you need to handle a number doable but annoying steps. Imagine tho that you could just download "The Site". Web server, blog app, database, everything. Including the OS. All as one (albeit large) file.
And then just upload it to a new host, where it Just Works. No reconfig because you're on a newer os or have to use a dfferent version of mysql. All of that is bundled up for you.
This is what virtualization (i.e. cloud computing) can buy you. And an open standard for infrastructure can make it easier to move your virtual system to other hosts.
http://www.theregister.co.uk/2010/04/28/nebula_goes_to_godda...
Does this mean they are completely moving away Eucalyptus?
we really want to make the cloud that hacker news users would want to deploy on their own infrastructure.
This reduces the risk that a competitor will come along with either (1) a popular vertically integrated solution that locks you out of market share or (2) some arrangement that allows them to dictate terms to you. Further, if a participant does emerge matching one of those patterns, they are organised respond through their established platform.
IBM does something similar with linux, which competes with non-free unix solutions and Windows. Google have done this with android, which reduces the risk of their being held to ransom by Apple.
Also, and again I apologize for not understanding your language here, but what do you mean by a "vertically integrated solution"? Could you give an example to illustrate what it would mean for a competitor to do this?
It would be interesting if there is a competing solution on the horizon, the success of which Rackspace is trying to preempt with this offering. I guess time will tell.
Swift is the storage platform for OpenStack. The compute platform will be released later.
