We had the chance last week to receive Rackspace in our London office for a technical talk on their technology offering and I wanted to share some thoughts and content on this presentation.

Rackspace has been around since 1996, playing first in the hosting space and moving on to the cloud space in the last years quite aggressively, with a lot of efforts put into OpenStack (among other). Based on the conversations and presentation, I wrote down a few elements worth keeping in mind about Rackspace’s service offering.

Hybrid co-located hosting

Rackspace offers a wide range of hosting services, from “traditional” hosting (i.e. anything from CoLocation to Managed CoLocation) to public IaaS cloud hosting in the Amazon fashion (public, accessible over the internet, multi-tenant, fully on-demand and pay-as-you-go pricing). The interesting part is that both service offerings run on the same stack (openStack) and can be hosted in the same datacenters, which means it opens the door to hybrid scenarios leveraging both types of hosting; the key advantage should be that, because of the physical proximity, no added latency between the two infrastructures should be introduced (contrary for instance to a scenario where an existing on-premise infrastructure would burst into a distant public cloud). Two scenarios to leverage this spring to mind in the context of Excelian’s clients in the Finance world.

For grid computing environments, where it is quite common to have three types of components involved in the infrastructure:

  • Compute nodes, running on fully commoditized hardware and potentially being virtualisation friendly.
  • Management nodes in charge of controlling the compute grid, running on high-end hardware and whose failure is handled by the architecture of the grid, but with potentially some impact on the infrastructure in case of failure.
  • Data grid nodes (Oracle Coherence for instance), whose storage nodes need to run bare-metal and on a specialised infrastructure (e.g. around a dedicated switch) and not being very virtualisation friendly
Leveraging Rackspace hybrid hosting, we could imagine installing the management nodes and the data grid nodes on bare-metal dedicated hardware (i.e. not shared with anyone) and install all the compute nodes on commodity cloud hosting. In case of a scenario where there is a need to guarrantee the availability of a minimum set of resources to guarantee SLA for an overnight batch, we could imagine an environment with 50% compute nodes running on the dedicated hosting (i.e. guaranteed) and the remaining 50% being provisioned on the cloud hosting based on demand.

For hadoop environments, a similar approach could be imagined with an HDFS storage cluster being run on dedicated hardware (which means allowing to store client-sensitive information for instance since it is running on an isolated client-specific environment) and then map/reduce computations running on cloud resources, scaling up/down according to compute requirements and manipulating confidential data only in memory, but without any storage in the cloud environment.These scenarios should be possible soon, once the support of openStack is official for both types of hosting.

The OpenStack model

OpenStack was launched in 2010 as an open source IaaS cloud management stack and is currently providing support for basic cloud services, namely Compute (Nova), Object Storage (Swift) and Image Service (Glance) which are the core functionalities of any cloud stack. The stack is virtualisation agnostic and supports most platforms, XEN being currently the better supported system. A new version of the stack is released every 6 months.

Rackspace is currently the main contributor to this open source project and the view behind this stack is to build an ecosystem of partners running public clouds on OpenStack. The objective is to leverage partners to improve the coverage of clouds and hence propose a cloud offering that would have a global reach much quickly than the competition. Currently, projections show that in the coming years there should be approximately 20 OpenStack clouds available across the world. DELL and Canonical are among the current contributors to OpenStack and will be offering public clouds based on the stack.

The presentation was concluded by an interesting conversation around the dynamics in the current cloud market and more specifically the role of Amazon: Amazon has been a leader and an innovator in the cloud space since the very early days and it seems that their strategy is to broaden as quickly as possible and as wide as possible the service offering in order to “thin” the market and potentially trigger consolidation around a few actors who can keep up with this arms race. To survive in this highly competitive market, it is vital today to have some differentiation factors like the ones mentioned above for Rackspace.
Andre Nedelcoux