EMC has been growing its profile in the IT sector, expanding beyond storage and obliquely into servers and software for some time. Now, it's getting ready to take a slightly more direct shot at the server biz with a forthcoming compute cloud offering under the Atmos brand, now in beta.
The company announced its Atmos storage cloud back in May, which is supposedly based on EMC's InfiniFlex 10000 disk arrays, which are made from x64-based controllers and SATA drives and which are known by their code-name, Hulk. It is unclear what iron is really behind the Atmos storage cloud.
It is equally unclear at this point what EMC has cooked up for its compute cloud, but the Atmos Online Compute Service has been outed by EMC's lab beta community, which explained that the service "allows you to self-service provision and manage servers within your virtualized data center in the cloud and use that environment to deploy and run your applications."
Basically, just as the Atmos storage cloud allows you to securely store data inside EMC's data centers, the Atmos compute cloud will allow you to run virtual servers and applications inside those same data centers.
You can read the "getting started" documents for the Atmos compute cloud here. As you can see, EMC has provided a framework that will allow complex companies to organize their virtual infrastructure, with virtual data centers for each organization, which in turn have multiple virtual machines and related applications and networks.
At the moment, during the beta for the Atmos compute cloud, each organization is being restricted to a single virtual data center, and is capped at 20 virtual machines, but the idea is to let companies deploy lots of VDCs and VMs.
The Atmos compute cloud will meter usage (and the fees charged for resources) based on GHz-hours for CPUs, GB-hours for main memory, and GB-days for VM image storage. Right now, the maximum VM image is capped at 20GB, but this will get larger as the Atmos compute cloud rolls out into production. EMC plans to offer three different VM sizes.
A small VM has a single virtual core rated at 1.5GHz and 2GB of main memory, a medium configuration has 6GHz aggregate (four virtual cores at 1.5GHz) and 8GB of memory, and a large configuration has 12 GHz of CPU (eight virtual 1.5GHz cores) and 16GB of main memory.
Given that EMC owns server hypervisor juggernaut VMware, you'd expect that EMC would create a compute cloud based on its own software, and indeed it has. The Atmos compute cloud supports VM images in ESX Server 3.5 formats - not the latest ESX Server 4.0, oddly enough, but more companies are on ESX Server 3.5 at this point than on the new release, so it makes sense for EMC to hold back while it is testing the Atmos compute cloud. Presumbly, when the compute cloud goes commercial, it will run on ESX Server 4.0 for cutting edge customers as well as on ESX Server 3.5 for customers not ready to move up quite yet.
The Atmos Online Compute Service will make use of a REST API to browse, run, stop, suspend, and update VMs running on the cloud, and will also allow the configuring of access to the Atmos storage cloud through NFS.
EMC has not said whose server iron it is running the Atmos compute cloud on, but it would be fitting, considering the ever-tightening relationship between EMC and server wannabe Cisco Systems that EMC would have drunk the California Unified Computing System kool-aid and built its cloud on the UCS b-Series blade and C-Series rack servers. It will be telling, in fact, if EMC does not do this.
GigaOm has reported that the Atmos compute cloud will launch this month, and that it has heard from sources that it will indeed be based on the UCS iron from Cisco.
EMC had not returned calls regarding the Atmos compute cloud at press time. You can take a look at the compute and storage clouds yourself here. ®