08 June 2015
Kieran Harty, CTO and co-founder of Tintri
In just 10 years the percentage of virtualised workloads has grown from 2 per cent to 75 per cent, according to a survey by Tintri. The rapid adoption of virtualisation has created a disconnect in your data centre. While your workloads are shifting to become virtual, your storage was built for a physical world. And so your storage is the source of growing costs, bottlenecks and frustration.
The most pronounced storage pain
In our State of Storage 2015 survey of 1,000 data centre professionals, the two most cited storage pain points were performance (50 per cent of respondents) and manageability (41 per cent of respondents).
That pain makes sense—the increased number of virtual workloads has generated far more random I/O patterns that choke disk-centric storage. To try and improve performance, storage admins tirelessly shuffle virtual machines from one storage LUN (logical unit number) or volume to another. Along the way they encounter all sorts of manageability shortcomings.
Flash – buying time
Most storage vendors are quick to recommend flash as a way to overcome performance pain. Flash is low latency and can handle random I/O. A single commodity SSD (Solid State Drive) is 400 times faster than a hard disk drive (HDD). To put that comparison in context, the speed of sound is “only” 250 times faster than walking.
But flash alone is a temporary solution—it only buys you time. Remember the root cause of storage pain is the disconnect between virtual workloads and physical-world storage; Flash does not address the root cause, it addresses the symptomatic pain.
Over time you will add more virtualised workloads and expand your footprint from virtualised desktops to servers to private cloud. To keep up with the pressure that puts on your infrastructure, you’ll need to keep buying more and more (high cost) flash. That’s good for your storage provider, but bad for your budget.
Better yet, you could resolve the disconnect once and for all.
VM-aware storage: buying a solution
To solve for the root cause of storage pain, you need to find storage that is specifically built for virtualised workloads. You need storage that is VM-aware. But what does VM-aware mean?
VM-aware storage has none of the remnants of physical storage – no LUNs or volumes, striping or widths. It operates at the most granular level, allowing you to take action on individual virtual machines.
That granularity is the solution to your performance pain. Conventional storage groups virtual machines into LUN or volume “containers”, and has you apply policies at the container level—assigning an amount of performance to be shared by the virtual machines inside. If there is a rogue virtual machine in the container, it greedily sops up performance that should be used by its container companions.
With VM-aware storage you can assign each individual virtual machine a specific performance level. That means you can give your mission critical application more performance, and put a cap on that rogue virtual machine. You can guarantee that every workload will get the exact performance it needs.
As for manageability, the ultimate marker is the spreadsheet. Most storage admins and/or virtualisation admins have to maintain a large spreadsheet that maps all virtual machines to their respective LUN or volume container. As those virtual machines are shuffled around, the spreadsheet must be meticulously maintained. VM-aware storage makes the spreadsheet obsolete. Login and you can see every individual virtual machine; drill in for full analytics or to set a policy (replication, cloning, etc.) at the VM-level.
Flash alone only buys time. Throwing more and more flash at the system does not solve anything. A storage admin can only improve performance in the long term with VM level manageability. Companies need to be able to see at the VM level to resolve the data centre disconnect.
The bottom-line… protect your bottom-line
You can resolve the disconnect in your data centre (virtualised workloads vs. physical-world storage), but you can’t do it with flash alone. The flash value prop is simple to understand and it buys you time by out-muscling random I/O; but as you add more workloads you’ll need to add more and more flash.
To address the root cause and close the gap, you need storage that specifically supports virtualisation; it must be fully VM-aware. With that degree of granularity and control you can solve for the performance and manageability pain that plagues conventional storage. That’s how you can save time and money, speed performance and realise your virtualisation vision.
