Amazon IaaS Compatibility

With the recent excitement about the Citrix decision to move CloudStack to the Apache Software Foundation, the IaaS providers find themselves jockeying for position. Most recently, Randy Bias of CloudScaling went on the offensive with an elaborate articulation of why CloudStack wasn’t the most Amazon compatible solution. Although Randy made several interesting points, the thing that really caught my attention was the actual analysis of compatibility. One comment posted the OpenStack compatibility matrix, see: http://wiki.openstack.org/Nova/APIFeatureComparison

The first thing I noted was the name on the three columns (Eucalyptus, CloudStack and OpenStack). It is said that a market exists when there are 3 good alternatives. Have we hit that point already? As a user of all three of these stacks, I’m comfortable saying that they work and each has gained enough market-share to continue gaining traction. Alternative solutions from Flexiant, Abiquo, Joyent, Nimbula, Morph, IBM/Platform Computing, etc. have not disappeared. However, I do feel that the solutions that aren’t getting listed in the comparison chart will need to find a niche. I do realize that this particular grid was created by OpenStack friendly people but even if that weren’t the case, it is the names that I see pop up most frequently.

The second item of interest was the use of ‘OpenStack’ as a category for comparison. To date, companies like Piston have leveraged OpenStack as their core and have built their extensions on the edges (installation routines, better dashboards, etc.) I suspect that the trend will quickly move from ‘competing at the edge’ to ‘competing at the core’. OpenStack implementations will need to differentiate based on substantive features. CloudScaling has indicated that their emphasis is ‘carrier grade’ solutions; this is a fine example of differentiation that will ultimately affect core architecture. Going forward, it will be less interesting to see if OpenStack has some feature – and more interesting to see whose implementation has the feature and the level of support for the non-functional requirements (massively scalable across geographies, ability to recover, etc.)

Finally, all of the solutions have made great progress in their goal to be AWS EC2 compatible – but in some ways they’ve all lost ground in recent months. I’ve given my sympathies to Marten Mickos (Eucalyptus), Jim Curry (Rackspace/OpenStack)and Sheng Liang (Cloud.com/Citrix) on the problems that they individually face as they build their solutions. These are complex platforms that require SIGNIFICANT testing cycles. Complicating matters, Amazon has unleashed major changes to EC2 in the last year around Virtual Private Clouds, Dedicated Instances, Reserved Instances, Spot Instances, new machine sizes, SSD offerings, modifications to clustered instances and so on. It will require an active community and deep pockets to maintain IaaS compatibility.

In some ways the IaaS wars have only begun. I can appreciate Randy’s position to ‘set the record straight’ in the press but ultimately I believe this won’t be a war fought in the press. It will be fought in production data centers. Stories will emerge on whose implementation works and whose doesn’t (and at what scale). The software will do the talking.

No comments yet.

Leave a Reply