Storage performance in the virtual world is a major challenge, outstripping the capabilities of many legacy arrays and leaving organizations with virtual environments that have not been able to keep up with growing workloads. The new all-flash and hybrid players in 2012 jumped in to fill some of the gap, and these companies have done well. Many of the storage companies that were “new” at VMworld 2011 and 2012 are now doing well on their own, each announcing double digit quarter-over-quarter gains.
This year, though, the focus has shifted to an emerging niche that has exploded. with multiple players jumping in over the past year with new kinds of solutions also intended to solve the storage performance equation in a less expensive and often mostly software-based way. The goal for these players is to allow customers to keep their existing legacy infrastructure intact while inserting just enough new hardware coupled with a software layer to help organizations push their gear and get more life out of it.
In essence, these solutions act as a server-side cache and leverage server-side hardware – solid-state storage and/or RAM – to work their magic. The major players include Infinio, PernixData, FlashSoft (acquired by SanDisk), and Proximal Data. It’s generally accepted that bringing data closer to compute provides the best opportunity for increasing performance.
Here is a high-level look at how these solutions operate:
While many solutions attempt to leverage flash to improve performance outcomes, a new entrant into this space – Infinio – takes an unusual approach, eliminating the need for solid-state storage and, instead, relying on host RAM as a caching mechanism. Infinio’s Accelerator product places a cache between the hypervisor and the physical storage layer.
Accelerator is a virtual appliance that is installed on each host. This appliance uses two vCPUs and 8 GB of RAM to create a local storage cache that doesn’t require any SSDs or other new hardware. It just uses existing RAM, and not very much at that. Installed on each host in the vSphere cluster, it creates an aggregated cache that is ultrafast compared to legacy physical storage devices. Whenever possible, Infinio satisfies read requests using either the local cache or cache from another host rather than requesting the information from storage.
Perhaps the best part of Infinio Accelerator is that it doesn’t require special hardware or the addition of any flash storage. It’s truly a software-only play and leverages just a little bit of existing host hardware – 2 vCPUs and 8 GB of RAM. Unlike some other solutions, Infinio doesn’t require the use of solid-state disks as the caching mechanism. It can be purchased with a credit card and downloaded and installed with no downtime or impact to running virtual machines. This ease of acquisition and deployment is one of Infinio’s differentiators.
At present, Infinio is a read accelerator only. It does not accelerate writes. But, given that most organizations read far more data than they write, and given Infinio’s relatively low price -- $499 per socket -- read acceleration can be a compelling opportunity. Infinio allows a free trial period, so trying out this product is almost a no-brainer for organizations that might be suffering from read-based performance issues.
PernixData sells what the company calls a Flash Hypervisor. The solution is a software layer that aggregates all of the solid-state storage from across all of the hosts in the cluster and uses it to create a scale-out data tier for the acceleration of primary storage. PernixData provides acceleration of both reads and writes across all hosts in the cluster. In addition, the solution can scale as organizations add more hosts or more server-side flash storage devices. Unlike Infinio’s Acclerator, PernixData’s FVP product operates as a hypervisor module rather than a virtual appliance. In some cases, this may mean that PernixData carries with it less latency than appliance-based solutions.
Beyond the flash storage it leverages, it requires no specialized hardware. As is the case with many solutions in this space, no major modification to the environment is necessary as long as the hardware exists on VMware’s hardware compatibility list.
PernixData’s solution implements a write-back cache solution, which carries with it some performance benefits. However, because data isn’t written to actual storage before it’s acknowledged as being written (if it did, that would be a write-through method), the potential for a host failure or some other hardware failure can trap data in an inconsistent state. PernixData has handily solved this issue by mirroring its cache to a second server. So, if one server fails, no data is trapped in the original cache. Storage guru Howard Marks goes into more detail here.
On the pricing front, the SMB Edition of PernixData FVP is priced at a flat $9,999. For this price, customers can accclerate the storage for up to 100 VMs across a maximum of four hosts with two processors and one flash drive each. The Enterprise Edition costs $7,500 per host, but it doesn’t cap the number of virtual machines, processors, or SSDs.
Proximal Data AutoCache
As is the case with PernixData, Proximal Data’s AutoCache solution also leverages existing server-side flash storage to acclerate storage performaance acess the cluster. The diagram below provides a broad look at what the solution accomplishes. In short, Proximal is able to cache hot data to local, host-based solid state disks or PCIe cards.¬¬
Proximal’s AutoCache tool provides customers with a full read cache, with write-through support to handle the write side of the I/O equation. In a write-through cache, writes are acknowledged only once they have hit primary storage. In the absence of other protection mechanisms, this is considered a safe method. If you want to learn more about caching, Fusion-IO has a great primer on the subject.
AutoCache is a hypervisor-level I/O connect tool that doesn’t rely on an appliance. Unlike PernixData, which works with any hardware on VMware’s HCL, Proximal has a more limited set of qualified hardware partners, but the company is continually testing and validating new hardware.
Pricing for AutoCache starts at $999 for flash cache sizes under 500GB.
Action Item: Many organizations have put major investment into their storage environments, but they still run into performance issues that can be very expensive to overcome with legacy storage vendors. These software-based solutions, particularly since they can be used on a trial basis, may help organizations solve their storage woes without having to rip and replace years worth of hardware purchases. CIOs who have architects considering forklift storage upgrades should insist that administrators give software-based tools a try first. In the worst possible case, the tool doesn’t provide a sufficient performance gain and needs to be uninstalled. In the best case, a CIO will be able to shelve the massive capex expense for another year or longer.