Vertical Scaling Ascendant - How are SSDs Changing  Architectures?
Wednesday, July 25, 2012 at 9:17AM
Todd Hoff

With Amazon announcing new High I/O 2TB SSD instances the age of SSD has almost arrived. I say almost because the $27K a year price tag for the hi1.4xlarge on demand instance type is outside the budget of many. Yet even at the full on demand rate the price per IOP for the high IO instance is attractive: 27 cents ($27K/100K IOPS) per vs $1.25 for disk. With the obvious benefits of giant SSD machines combined with 10 Gbps networking, it’s interesting to consider: what architecture decisions might you make differently in the future?

More Headroom for Vertical Scaling Simplifies Everything

The beauty of higher hardware performance is it shifts effort away from the programmer which allows developers to focus on the business of business, minimizing trickeration. This has always been the allure of vertical scaling and is well realized by SSDs through a combination of high throughput, low latencies, and just as important, high densities.

We have a few early examples showing the performance punch of the new High IO instance:

But not all is bliss in the garden. Jeff Darcy wrote in Testing [GlusterFS] on Amazon’s New SSD Instances  that they’ve found real world performance unimpressive and even worse in some cases, concluding:

They’re certainly a welcome improvement for this worst-case kind of workload, but I’ve seen their ilk before so the only thing that’s really new to me is the high price tag.

Practical Implications of the High IO Instances

Fewer Machines Means a Lower Total Overall Cost

High Availability is Still Necessary

It’s Not All Peaches and Cream

Related Articles

Article originally appeared on (
See website for complete article licensing information.