The Economics of Cartesian Scaling
by David Merrill on February 25, 2010
Can you think and talk in 3-dimensions? Our theater-going experience seems to have taken on a 3-D standard as of late; perhaps our operational, technical and economic metrics around storage is about to do the same.
Hu posted a recent blog on Cartesian methods to describe scale-up, scale-out AND the virtualization dimension or ‘scale-virtual.’ This idea presents some new options to define the economic, performance and extensibility of a particular storage system and storage architecture.
- X-axis is the scale-out dimension. Measurements are well understood in terms of $/GB or costs associated with expansion.
- The Y-axis is the performance view. With a rapid increase in virtual hosts connected to storage systems now, performance (more than capacity) is the key area to consider in the design. $/IOPS is a typical measurement to show the costs to achieve a certain performance rate. Adding ports, cache, more spindles and better RAID levels can be applied to meet this rate
- The new Z-axis can describe the extensibility of capacity and asset utilization available through virtualization. This is not scale up or out, but scale-virtualized. The key metric for this scale is return on asset (ROA), being able to leverage existing capacity to extend the size and effectiveness of a virtual storage pool.
I am not really fluent in 3-dimensional conversations, but there seems to be a new set of metrics around $/IOPS/GB as well as the ROA view. SSD technology introduces the $/GB/IOP metric, but this needs to be taken further.
A new set of common metrics using this 3-axis or Cartesian system can certainly help with the design and planning of highly scalable cloud architectures. It will also help normalize orangapples to applorange comparisons…… since so many new storage systems have multi-dimensional capability.
Comments (1)
Michael Hay on 25 Feb 2010 at 5:43 pm
David, this is concise and thought provoking post! Thank you! Perhaps we can start from the origins of the Cartesian scaling concept: super computers. There the X and Y axes are very similar, but the Z axis might need a little looking into as well. Perhaps we can take a lead from the PERCS proposal that IBM put forward winning the Blue Waters project to build at 20 Peta-FLOP super computer in Illinois. Here the P is not for performance, but productivity which in a sense is a kind of return on asset.
