Five Rules for Choosing an All-Flash Array for VSphere Workloads

Flash technology is newer and faster than disk storage, and is changing everything about storage for virtualizing your datacenter. But buying flash enabled storage for virtualization is different than buying traditional storage. Chuck Dubuque, from Tintri explained the five rules for choosing an All-Flash Array storage for enterprise datacenter.

In my last blog, I’ve just explained what actually the All-Flash Array technology is? In this blog, I’m going to share 5 rules for choosing an All-Flash Array Storage for your enterprise datacenter.

Five rules for choosing an All-Flash Array storage are as follow.

  1. Performance is controlled by Controller

Traditional array of disks provides more capacity but it also gives more spindles which needs more performance. In enterprise datacenters or cloud setups, they have lot of racks with full of storage devices and you need huge performance to access your data from different disks. In flash array disks, the controller is everything. An individual SSD can deliver between 50 – 100K IOPS in a single disk, if you got an array of 24 – 32 disks, your SSDs are capable of delivering many more IOPS than most controllers are made for today. So, the controllers are the main items for delivering the performance.

  1. Data Reduction Technology

When deploying flash technology in datacenter, the most efficient mechanism is to adopt data reduction inline. Data reduction techniques (Data Reduction = Compression X Deduplication + Thin-provisioning) such as compression, deduplication, and thin-provisioning on spinning disks are at high cost in the form of performance and latency. While flash disks are so fast as cloning, link cloning, so adapting data reduction technologies are 3-10X faster as compare to spinning disks depending on different vendor architecture and reduced the cost in the form of better performance (clones, linked-clones, thin-provisioning), and low latency.

inline deduplication and compression in our point of view are mandatory, and thin-provisioning benefits you get the most by how well integrated with VSphere or other hypervisor. From performance perspective, inline compression, inline deduplication is happened in real time before anything actually written on flash drives and your controller is spending compute cycles to compress and dedupe and write those very large blocks on flash disks to save physical space. Performing all three data reduction methods (compression, deduplication, and thin-provisioning) in which data is written on flash drives is called inline data efficiency.

 3. Integration with Hypervisor

 In a virtual server environment, the interaction between hypervisor and the storage hardware that supports it is complicated. In an effort to simplify that interaction and make it more efficient, VMware developed the vStorage APIs for Array Integration (VAAI). vStorage API for Array Integration (VAAI) is an application program interface (API) framework from VMware that enables certain storage tasks, such as thin provisioning, to be offloaded from the VMware server virtualization hardware to the storage array. The APIs create a separation of duty between the hypervisor and its storage devices, enabling each to focus on what it does best: virtualization-related tasks for the hypervisor and storage-related tasks for the storage arrays.

With VAAI, storage array vendors can directly integrate their storage hardware and applications with vSphere. VAAI enables certain storage tasks, such as cloning, to be offloaded to the storage array, which can complete them more efficiently than the host can. Rather than use host resources to perform the work (which was required prior to VAAI), the host can simply pass the task onto the storage array, which will perform it while the host monitors the progress of the task. The storage array is purposely built to perform storage tasks and can complete requests much faster than the host can.

Vendor Support for VAAI

Currently, the vStorage APIs for Array Integration provide benefits only for block-based storage arrays (Fibre Channel or iSCSI) and do not support NFS storage. Vendor support for VAAI has been varied, with some vendors, such as EMC, embracing it right away and other vendors taking longer to integrate it into all their storage array models. To find out which storage arrays support specific vStorage API features, you can check the VMware Compatibility Guide for storage/SANs.

EMC Scale IO Model Details

  1. Commodity Hardware Increases Productivity

Flash was appeared in enterprise for virtualization either in the form of SSDs which actually run on local host which did not give you enterprise storage but give you very nice performance boost or PCIe card base Fusion IO which provides the feature of iSCSI with great raw speed in enterprise datacenters. For selecting a flash storage with hardware for virtualized workloads, two approaches are being used, i. commodity flash hardware, ii. Custom flash hardware.

i. Commodity Flash Hardware

A year or two ago price difference between flash and spinning disks was 2:1 ratio, and this is a very huge difference for enterprise or scale-out systems. When flash was introduced, its price was too high about $5/GB, but in these days its prices are decreasing and now its price is around $2/GB. This is all being driven by the hardware available off-the-shelf without any modification. So this becomes very broadly applicable to all of your workloads. These kind of transits really driving us within a few years to flash will be fairly ubiquitous for virtualized workloads. So, in our recommendations for server virtualization, VDI, cloud applications, and cloud storage models like object and file storages, commodity flash hardware can be the best way to go. You can get your best bank for your buck you are getting 99% of performance of custom hardware at much lower cost, much higher trajectory of cost savings density increases the power and cooling savings.

ii. Custom Flash Hardware

Custom flash hardware still has its place, there is a lot of places in high performance computing (HPC), big data modeling, custom high speed real time kernel workloads, where a custom flash solution is really good fit, but ROI must be clear. Because a custom flash solutions come at a high price premium and also low your degree of flexibility in terms of future upgrades.

Whether you choose the platform flash, soon price shouldn’t be an object. One of the thing in price including packaging, power, cooling, maintenance, space, and disk reduction. So, these are the numbers including disk compression, inline deduplication, and from thin provisioning, and other software on top of the flash, because flash is so fast, you can now more innovate of things. And if you buy flash storage for enterprise from any vendor, you should insist for those features should have.

  1. Flash Speed Enable Software-defined innovation

Flash really gives you the ability to do more interesting things, VM-aware Aware Storage (VAS) infrastructure really gives you some benefits, so you can look for new features that are layered on top of flash that give you: –

i. Guaranteed VM performance and VM-level QoS and measurement which is much easier with flash as instantaneous access of data because flash is so fast so, you should be getting more data about your environment from a flash array.

ii. Real-time actionable VM-level analytics as your system is designed around disks based operating systems, the data you are getting may not be useful as its reporting real-time on the performance of VMs. So, you look for things that can give you VM-lever analytics either real-time or after the fact and of course.

iii. integration with Hypervisor really key to getting the best bank for your buck from flash because of the additional saving you can get really tight integration with thin-provisioning, cloning, linked-clones, and data management functionality that integration with hypervisor.

Modern flash arrays can consistently deliver 100,000 IOPS with latencies below 1ms. If you’re considering deploying a VDI at your company and want to support several thousand users, an all-flash array can erase your fears of boot storms destroying morning productivity. The low latencies flash delivers can get everyone up and running without any detectable performance impact. Not only that, but you can reduce storage provisioning for virtual desktops and servers from several minutes to 50 seconds or less. You can update thousands of virtual desktop images while your coffee’s brewing rather than having it take up your whole lunch break.

Just a few years ago, flash storage was just a way to boost the performance of hard drives and as a tool for individual consumers to speed up their computers. Now, however, businesses are able to use all-flash arrays to reduce latency and increase IO all without spending more than they would for hard drive arrays. It’s hard to doubt the utility of flash storage and we’ve reached a tipping point where hard drives will finally fall out of favor.

Leave a Reply