Since when did software-defined mean that hardware doesn’t matter?
- 01 June, 2020 15:55
It’s a question that’s worth unpacking because the way we see it - hardware is at the heart of a performant data centre, yet it seems to be an afterthought in today’s complex storage environments. ‘Software-defined everything” promises the world: flexibility, scalability, centralisation, and reduced capital expenditure - so where’s the advantage in scrutinising the hardware? What benefit can it bring to the results we’re seeking? And how does it translate to real, tangible value for end-users?
Hardware has been commoditised in the data centre to the point of obscurity, and in so doing, we’ve taken a giant step backward because it’s been to the detriment of the results we’re seeking. We’ve been led to believe that you can’t have it all and you need to choose between performance, efficiency, and cost. At best, you can only have two out of the three.
Picture a pendulum
This has led to two extremes, so imagine a pendulum: on the one side you have a proprietary, on-premise solution. Yes, this allows a high degree of control and it resolves a lot of the overhead but you’re locked in and pay for it again and again… On the other side it’s a D.I.Y. setup with open source software (OSS) and commodity hardware. After the initial cost of the cheap hardware, the ongoing cost is free! Except those who have struggled with the complexity of such a solution will tell you: it’s not.
Generic is the enemy of progress
In either scenario, it’s not unusual to see half-empty racks in data centres these days. Why? Because generic appliances are assembled using off the shelf components that are not optimised for the task at hand. We’ve been going about it backwards: shoe-horning software in an attempt to force better results. The irony is that in reality, the net result is large, inefficient, power-hungry boxes that suck power and cooling resources. As data continues to grow and the trend to hybrid cloud sees ever more data repatriated from the public cloud, this is a situation that - left unchanged - is unsustainable.
Specialisation fosters performance breakthroughs
There is a better way. It’s where the pendulum comes to rest, and it’s what we call “task-specific” hardware, purpose-built for the data centre and engineered from the ground up to exploit the capabilities of leading OSS for the data centre across storage, networking, and compute.
We’ve demonstrated that when you use task-specific hardware, optimised to run the very best software-defined, open-source solutions like Ceph for storage or SONiC for networking, you can achieve significant and meaningful gains. And, especially when it comes to storage; density improvements that enable you to fully utilise available space without blowing power and cooling budgets.
For example, SoftIron’s HyperDrive Density Storage appliance enables you to deliver 120TB of storage in a single “u” of rack space with a power budget of fewer than 125 watts – that’s over 5PB of storage in a 42U rack, delivered for around 5KW of power consumption – well within that delivered in most data center racks – and a potentially significant cost saving in co-location environments where consumption impacts rack rates.
Of course, it’s not just about power, it’s also about performance. When you truly understand the task and the way the software is architected to deal with it, you can exploit the potential with hardware that is custom-built to support that. We dive deeper into that here, but in an essence, that’s why we’re able to deliver such blistering performance using open source code. Furthermore, when you design and build hardware for a specific task, and you know exactly what code you’re going to run, you can vastly improve the ability to efficiently and intuitively configure, install and operate your environment.
Everyone’s a winner
The end result? All the benefits of a proprietary bundled solution, but without the huge drawback of the vendor lock-in because don’t forget, it’s still open-source, so you can move on at any time.
If you're selling generic hardware into open-source data centre deployments you're missing a trick. You’re missing the chance to differentiate by delivering superior performance and efficiency, as well a solution that's easier (and therefore more profitable) to deploy and support, yet still delivers all the independence of open source.
You can certainly continue to run - or sell - Ceph, SONiC, or any other software-defined platforms on generic hardware. Our bet though, is that once you’ve experienced the difference that purpose-built, task-specific design makes, you won’t want to.