Horizontal Scaling Alone…
by Michael Hay on Dec 5, 2009
Makes about as much sense as IP over avian carriers. Seriously if I look on the Top500 super computer list I can see an emerging trend towards hybrid/many core computing with the combination of vertical and horizontal scaling — sometimes called Cartesian scaling or scale-up and scale-out. There is still a lingering hangover in the IT industry about massive cheap scale-out only architectures. One of the original innovators of this approach was NASA Goddard with their Beowulf concept. From there scale-out only architectures have become somewhat of a fad in my opinion. That is until these systems ran into issues related to power efficiency, density, heat production, etc. I would say that being more Green is proving to be an unstoppable force to be reckoned with and even Beowulf cannot slay this dragon.
At the end of November Hitachi GA’ed the Hitachi Content Platform V3. Like my colleague Claus, I will not focus on a lot of the details of the announcement, I will zero in on one point: Cartesian scaling. HCP V3 supports 96 nodes in a single system image, with value added features like capacity reduction technologies, integrity checks, namespaces, etc. While the raw number of nodes might raise some eyebrows, I believe that is because of the fad of one dimensional, scale-out only architectures. The fact is that HCP can scale to an order of magnitude more storage capacity than other products on the market today with just 96 nodes. The primary reason for HCP’s ability to scale comes from the fact that we invest in Cartesian scaling. Here are the basic approaches:
- HCP V3 model 300 – RAIN configuration with each node being symmetric in nature scaling up to 85TB of storage capacity across a total of less than 30 nodes (each node has a quad core Intel Xeon inside and 4GB of main memory)
- HCP V3 model 500 – SAIN (SAN + Array of Independent Nodes) configuration which can be bundled in an appliance configuration with our AMS2000 series, or as a diskless approach to attach to an existing storage infrastructure. In this model of course each node is more powerful (8 cores and 12GB of main memory) they are also symmetric, but the most important part is that the nodes do not contain internal HDDs for user data. Instead we rely on best of breed Hitachi storage to do the trick. This allows for improved capacity efficiency as well as the ability to add storage to an existing node. In effect this lets us scale storage capacity without having to scale compute at the same time. Here we support up to 40PB of storage or 10x more than the nearest competitor
As a premier provider of storage infrastructures Hitachi prides itself in in consistently investing in all technologies and architectures which make our products best of breed. We consider investments in scaling-up and scaling-out leading to offerings which are naturally able to slay the Green dragon. Examples of systems where Hitachi has made an investment in Cartesian scaling beyond HCP includes USP-V, HNAS, HENP, and HDDS.
Comments (2 )
[...] This post was mentioned on Twitter by Hitachi Data Systems, Jessica Longly – CDW. Jessica Longly – CDW said: RT @HDScorp: New post: Horizontal Scaling Alone… http://bit.ly/5A8xRw [...]
[...] fellow HDS bloggers Ken Wood and Michael Hay have been using the term Cartesian scaling to describe what I call Scale up and Scale out. Cartesian [...]