Virtualising servers and desktops is top IT priority for many organisations right now – and with virtualisation comes the need to provide virtual server and virtual desktop infrastructure (VDI) storage to support it.
But virtual server and VDI storage is not like storage in a purely physical environment. Most people opt for shared storage and choices need to be made over which storage protocols to use. Meanwhile, storage performance becomes critical and storage management tasks are no longer solely carried out on the storage controller.
In this interview Cliff Saran, managing editor (technology) at ComputerWeekly.com, talks with Antony Adshead, storage editor at ComputerWeekly.com about the three things you need to know about storage for virtualisation and VDI storage.
Cliff Saran: Antony, server and desktop virtualisation are top of many peoples’ to-do lists right now. But what key things do they need to know about virtual server and VDI storage?
Antony Adshead: Well, the first thing is that you need shared storage, a storage array; SAN or NAS – and in some ways that’ll make things simpler than storage directly attached to every server. Apart from the fact that directly-attached storage is wasteful, shared storage is required to take advantage of the features of server virtualisation – such as the ability to easily migrate virtual machines and reassign storage between them. It means you are not bound by siloed and relatively low capacity drives, and can take advantage of easier to manage methods of increasing scale, adding performance and providing resiliency.
Having said that, you will face a number of choices. The first is whether that shared storage should be block or file access. There is no overall advantage to either. At a basic level, file access (ie NAS) is less costly and simpler to implement. It sits on the LAN and needs no special skills. Block level access (ie SAN), with iSCSI connectivity – which also can operate on the existing LAN – may require greater networking knowledge than a smaller organisation has in-house and Fibre Channel definitely requires specific in-depth skills.
Beyond that, NAS’s NFS and CIFS file systems are easier to mount and configure than iSCSI or Fibre Channel, but they lack the advanced features, such as traffic queuing, priority and resource management that the block protocols have. NAS suppliers have, however, introduced support for cloning features such as VMware’s Linked Clones and support for data de-duplication – both of which reduce capacity requirements and make NAS a contender for file storage in virtual environments.
SAN block-level storage has key advantages in terms of performance, not least because you’ll be mapping storage directly to virtual machines in what VMware calls raw device mapping – other suppliers have their versions too – and this allows advanced features like clustering between virtual (and physical) servers.
Saran: Excellent. Thanks Antony. What’s the next thing to be aware of?
Adshead: Well, with virtualisation, storage performance becomes much more critical than before.
That’s because, whereas before each server – probably with one application on it – had access to one set of on-board storage media, now numerous virtual servers or desktops are attempting to access the drives on your shared storage. Because there are multiple servers all operating at once, with differing workload profiles, the resulting I/O and throughput requirements are irregular and random, especially in VDI environments, which experience so-called boot storms when users log on at the start of shifts.
So, to optimise performance on shared storage you are going to need to look at techniques such as wide striping and data tiering. Wide striping brings many RAID groups on a SAN into one pool and stripes data across multiple drives, so that blocks are being accessed by many read heads simultaneously, so cutting read times.
Data tiering sees the placement of data on drives of different performance levels according to how frequently used it is. Automated data tiering is available on many SAN products and some NAS devices.
The key bottleneck for storage in any virtualised environment is read speed, so here we have seen solid state drives enter the fray to alleviate the problem. You can add them into standard drive bays on a storage array, or buy dedicated all-flash arrays.
Also, lately, we are seeing interesting new array products on the market that target virtualised environments and combine fast flash storage with cheaper, spinning disk and algorithms that reduce data to boost capacity via data de-duplication or similar techniques. These products come from the likes of Nutanix, Tintri, Tegile etc.
You’ll also need to pay special attention to capacity planning. Virtual machines can spawn at an incredible rate and you don’t want to be caught with your pants down, capacity-wise. So, you need to be analysing, profiling and predicting the storage needs of your virtual servers and there are a number of tools available to do that.
Saran: And what’s the third thing, Antony?
Adshead: The third key thing to know is that storage is no longer something you manage separately from servers; now virtual server environments allow a number of management functions to be carried out from the hypervisor.
VMware’s hypervisor, vSphere, for example, can use APIs that link the virtual machine environment to the storage array. This offers the ability to manage storage performance from the hypervisor (using vStorage APIs for Array Integration) and discover, monitor and manage storage in the most appropriate way for your virtual machines (using vStorage APIs for Storage Awareness). There are also plugins available from most of the storage array suppliers that allow admins to create LUNs and clone and snapshot virtual machines from the array.
So, those are the three key things you need to know about storage for virtualisation; you’ll need shared storage; storage performance is critical, and; storage can be managed from the hypervisor.