Cloud computing is creating significant shifts in the way information technology is provisioned and consumed. This also kicks off a challenge to cloud service providers find innovative ways to adhere to the high demands of availability, performance, efficiency, and cost-effectiveness.
AMD’s SeaMicro SM15000™ brings together compute, storage, and networking very effectively to solve huge costs and space issues with hardware required to operate a cloud. This paper describes the SeaMicro fabric, architecture, and tips to realize a compute cloud using Rackspace Private Cloud Software (RPCS).
The AMD SeaMicro SM15000™ brings compute, networking, and a super-compute fabric together in a single 10 Rack Unit (RU) chassis. It may be configured with the following:
A total raw capacity of 6,848 TB or 6.69 PB may be attached to the system, including both internal and external drives.
An interesting and powerful aspect of the SM15000™ is the supercompute fabric, which is able to deliver 1.28 Terabits/second of bandwidth internally. It is configured as a three-dimensional torus, tunneling packets using flow control digits (FLITs) and routing them through a “wormhole fabric”. The “wormhole fabric” uses a simple protocol to route packets based on whether virtual channels are active, waiting, or idle. In cases of high congestion, the traffic is prioritized using virtual channels. Storage traffic is always given the highest priority to avoid interruption.
OpenStack Compute, code named Nova, is the cloud operating system to launch and organize a compute cloud, which includes creating and running instances as well as managing the networks between virtual machines.
OpenStack compute consists of different services, but they are mainly categorized into two types of nodes:
The API service is responsible for receiving and responding to the user request. Once the request is received, the nova-scheduler service finds the right compute host and sends it a message. All of these services talk with each other using a RabbitMQ that is sitting on the controller node. MySQL database that is also sitting on the controller node is the central database used to store any persistent data.
You can interact with the APIs using OpenStack python-bindings or you can use Horizon, which is a graphical user interface to interact with the cloud.
AMD SM15000™ brings a good balance between storage, compute, and networking. The SM15000™ comes with 64 compute cards, 64 internal storage cards that can be extended up to 2 TB per card, and 16 ten Gigabit Ethernet (GbE) in single chassis that makes this environment very attractive. A recommended architecture for the different compute services running in the SeaMicro fabric is as follows:
The SM15000™ incorporates a technology to interconnect all the nodes including computing cards, disks, network-interface cards, power supplies, and fans. With this technology, the SM15000™ provides multiple paths between any two components in the system, which allows system for easy routing during node failures.
Each SeaMicro fabric comes with 6 slots for AC/DC power supplies and 4 optional slots. Failure of a single power supply doesn’t have any impact on the system availability and operation.
Installation of RPCS is well documented by the following article:
AMD SM15000™ provides an efficient combination of compute, storage, and networking in a compact box and is a strong choice for creating compute cloud in a box. It consumes less space and power compared to other solutions, and is just one system to manage.
© 2011-2013 Rackspace US, Inc.
Except where otherwise noted, content on this site is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License