It’s a question that used to refer to investments or supply-chain management, but it’s now the latest question being asked of IT strategists and architects. What is your commodity strategy?
Today’s enterprises are taking a page from the infrastructure playbook of web giants like Facebook, Amazon, Netflix, and Google. The FANG companies and just about every successful cloud-based business build hyperscale data centers atop inexpensive servers built from readily available components. The nature of their business has led them to eschew traditional infrastructures in favor of commodity-based, distributed systems architectures. It would simply cost too much to house every byte on an EMC array and process every bit with an IBM Power server. By building and using software that harnesses the power of lots of independent, low-cost nodes, web-scale companies keep costs down and successfully serve millions of customers.
And the enterprise wants in on the game.
Commodity ≠ Cheap
An article posted by Rich Castagna with TechTarget this week criticizes software-defined storage vendors for pushing “commodity." I get his point. You don’t want to trust your storage to crap hardware. Before joining Hedvig – as an ex EMC-guy – I too had a perception that commodity meant “cheap and built cheap” – but I’ve since seen the light.
Wikipedia defines commodity as: an economic good or service when the demand for it has no qualitative differentiation across a market. This is really where we are at with the server market. The quality of a given commodity node may differ slightly, but it is essentially uniform across producers.
At Hedvig, we very much believe that hardware still matters. Yes we use phrases like whitebox server, off-the-shelf hardware, and commodity nodes – but we don’t suggest that you go to Fry’s, grab a cart, and load up (though I have picked up a few really nice Supermicro servers at a local shop called UNIX surplus – the home of the $99 server – no joke.) We are promoting the idea that you can build a great enterprise infrastructure from quality, standard hardware with standard x86 (or ARM) processors, memory, flash/SSD, and hard drives – components that are quite similar if not identical to those you will find behind the bezel of today’s trusted enterprise storage hardware solutions.
The magic in the software
The commodity strategy question is a two-sided coin. On one side is the hardware equation, on the other is software. Much of the value in today’s infrastructure is shifting to software vs. the silicon and circuits it runs upon. The software matters. What the web-scale guys do is create and deploy software created for distributed computing environments. One example is Apache Cassandra, invented by Hedvig founder Avinash Lakshman when he worked at Facebook. Distributed systems software is designed for hardware failure. It knows how to deal with the inevitable downed server, dead hard drive, and fried NIC. The expectation is that these failures WILL occur in the hardware layer. With a commodity approach, a component or node can fail and you can replace it with any like component, even of a different brand, and the system picks right back up. In fact, you could even have a slightly different component – maybe a bit bigger, or a bit faster – and the system doesn’t blip. Redundancy and the ability to self-adjust and repair are key functions of the software code. NoSQL, Hadoop and SDS systems like Hedvig are built this way.
There are significant benefits to the commodity-based distributed systems approach:
- Lets you deploy incrementally and dynamically as load or capacity increases (see LKAB case study for a customer perspective).
- Lets you purchase standard server “building blocks” that can be utilized for one or more services – no more islands of single purpose hardware.
- Lets you ride Moore’s Law, always sourcing components at the best rates available and taking advantage of hardware advances.
- Lets you phase systems in and out non-disruptively like a conveyor belt that over time drops off the old and brings on the new (look ma, no forklift upgrade!).
There are right approaches and wrong approaches. The type of nodes you deploy and what's in them will determine things like IOPS performance, and the quantity of nodes will determine your availability. Done right, a sound commodity strategy will give you what you need to be successful and help you lower the costs of owning and operating a data center – as much as 60% or more according to Gartner. Here’s a good reference article by Drew Robb on Enterprise Storage Forum on some dos and don'ts.
Server hardware has reached a really impressive price / performance / quality level. You don’t have to spend tons of time and effort to build a relationship with an original design manufacturer (ODM) to be successful. You can buy really nice systems from companies like Cisco, HP, Dell, Supermicro, and Quanta at really a decent price. And with more and more software solutions being built to take advantage of a hyperscale, distributed systems approach, you have lots of application choices to drive your business effectively in the new world.
So – what is your commodity strategy? If you don’t have one, maybe now is the time.
To learn more download our free eBook: Hyperscale Storage for Dummies.