Keeping control of your storage environment: whatever the weather

Over the past couple of years, many enterprises have experienced the painful mismatch between traditional storage and the virtualised data centre. While virtualisation had brought many great benefits – it has come at the cost of application and VM performance. Application performance is the number one issue organizations face when they virtualise their servers. The main cause of this is what has become known as the “IO Blender” – where optimised IO leaving a VM is blended and mashed together with 10, 20 or even hundreds of VMs, to create a highly random IO pattern that traditional storage is not capable of dealing with very well. VM-Aware storage and VM-Optimized storage are two ways of dealing with this problem. But there are significant differences with the results of each method.

VM-aware storage

VM-Aware storage taps into the fact that VMWare can use NFS as a datastore. VM-Aware storage implements an NFS server that presents a single namespace as a VM datastore and VMFS uses files as logical containers for each and VMs VHDs.

It’s simple, clean and most importantly – provides visibility into which VMs are consuming what storage resources. Now you can see exactly how much storage or IOPS a specific VM is using. Having that visibility solves part of the performance problem. If a particular VM starts to consume too many IOPS, impacting the other VMs, now you have visibility into the culprit and you can take action. Maybe move it to another host, or restart the job at a later
time when the rest of the system is not
busy or kill the runaway process on that
VM. VM-aware storage gives you visibility
into how storage is being utilized and by whom.

While visibility is a major step forward, the major limitation of VM-aware storage is that it cannot address the IO Blender problem. This is a little bit like telling you the weather from yesterday – useful, but it doesn’t predict the weather tomorrow and it can’t change it today.

This limitation comes from the fact that VM-aware storage is still a storage array presenting an NFS server. The NFS protocol was born in 1984 and designed for file sharing – this is practically 20 years before the advent of virtualization. The NFS protocol has no concept of VMs, and it cannot de-blend the IO originating in the hypervisor – the IO simply ends up in logical containers (files) that allows you to report on what resources are consumed per file (per VM). The NFS protocol is still the NFS protocol and the highly random IO produced by the hypervisor IO blender still piles up at the end of the network connection where the NFS server is implemented. You may have visibility, but the IO is still random and without priority.

VM-optimised storage

The modern alternative is VM-optimised storage. VM-optimised storage is a Software-Defined Storage architecture that is built specifically for virtualisation.VM-optimised storage places a software based virtual storage controller that operates inside the hypervisor to eliminate the IO Blender at the source.

By operating in the hypervisor, VM-optimised creates a virtual storage stack that isolates IO for each VM and stops it from blending before it leaves. This isolated IO is then dynamically optimized (with no manual intervention or tuning) according to the IO pattern of the VM. With the IO isolated and optimized per VM, IO can now be prioritised per VM to guarantee that the most important VMs get the resource they require. VM-optimised storage is the only storage that delivers per VM visibility, optimisation and prioritisation of IO end-to-end. No more IO Blender – just high performance IO per VM with complete control. VM-optimised storage delivers the visibility of VM-aware storage and uses a storage architecture that is purpose built for virtualisation to add per VM IO control to both accelerate it and prioritise it. End users can see and control exactly what storage resources are being consumed by each VM and their environment. Put another way, instead of just telling you what the weather was yesterday, you’re now in control of the weather in your environment.

Optimised storage for Hyper-V

Gridstore is software-defined storage that is optimised for Hyper-V. It is purpose built to accelerate applications in virtualised environments. Patented Server-side Virtual Controller™ Technology (vController) dynamically optimises and prioritises I/O before it leaves the server and leverages massively parallel network connections to accelerate I/O to massive scales.

The following capabilities, developed by Gridstore, are all designed to deliver accelerated application performance.

vmOptimized™ Storage

This is designed to dynamically accelerate
all virtualised applications. Every application has an optimal I/O pattern – Gridstore delivers it. Gridstore vControllers operate on the server and create a virtual storage stack for each VM. vControllers isolate I/O from each VM and dynamically optimize this based on its I/O pattern – without any manual tuning. By eliminating the I/O blender at the source, Gridstore can deliver optimal I/O for each VM and accelerate all applications.

Gridstore accelerates I/O for every application and eliminates the cost of over provisioning storage to compensate for the I/O blender.

TrueQoS™

This is designed to guarantee performance for your most important applications. Not all VMs are equal. Unlike storage arrays that can only implement QoS inside the array (where it is too late to prioritize), TrueQoS leverages server-side virtual controllers to prioritize I/O before it leaves the server to eliminate the performance impact of “noisy neighbours” on your most important applications.

TrueQoS is policy driven and set by your hypervisor. TrueQoS is the industry’s first end-to-end storage QoS and works across all Hyper-V hosts as well as physical servers in your data centre. Quite often, the most important workloads are not virtualised, now these can be prioritised and managed seamlessly with TrueQoS.

GridScale™

Massively parallel performance scaling

GridScale is a simple, pay-as-you-scale building block approach to storage. Now end users can easily grow both capacity and performance in line with their budget and storage needs. Unlike traditional storage that grows capacity behind a single controller, GridScale enables storage to grow more powerful as you add storage nodes. All this combined delivers massively parallel performance.

All storage nodes are virtualised into pools (vPool) and managed seamlessly through a single pane of glass. Each vPool can contain up to 250 storage nodes allowing to scale from 12TB to 3PB of capacity with 1,000 Xeon cores, 8TB RAM, 5TB bandwidth and 100TB of PCIe Flash Cache. That’s GridScale.

Grid Building Blocks™

Intelligent storage nodes built on x86 hardware

Gridstore is delivered in a 1U storage appliance that is attached to standard 1Gb or 10Gb Ethernet. Each storage node contains all required software with unlimited vControllers and management – the only thing added is annual support. Gridstore Storage Nodes come in two types:

C-Class – designed for capacity needs with high performance streaming I/O

H-Class – with PCIe flash cache plus SATA to satisfy IOPS intensive workloads without breaking the budget

Overcoming traditional storage obstacles

Winthrop University is a nationally accredited US university that offers undergraduate, graduate and continuing professional education programs of national caliber while also focusing on student development and engagement. Founded in 1886, the university today is attended by 6,170 students. Winthrop University has a centralised IT infrastructure that revolves around supporting its mission-critical ERP system, which is used for everything from meal planning and student ID cards to housing programs and academic computing labs. As the technologically advanced university continued its growth, storage became a focal issue.

The university manages a centralised storage infrastructure which covers over 100 Windows and Linux servers. Using a mix of DAS, NAS and SAN storage, the college was quickly finding that storage limitations were impacting application performance, reliability and even worse, flexibility for future expansion. Its DAS solutions were inflexible, NAS solutions became inefficient to scale with islands of storage to manage, and SAN solutions costs were skyrocketing.

“We found that our existing storage infrastructure was becoming non-scalable and very limiting,” says James Hammond, associate vice president for information technology at Winthrop University. “Our DAS environment was extremely inflexible and our various NAS islands were proliferating, making it difficult to manage and expensive to scale. We thought SANs would be the silver bullet that could solve our challenge, but the price tag was very painful. We were also very frustrated because the return on investment for the SAN was out of sight due to the high cost of storage and management complexity. We knew there had to be a better solution for the long term.”

Better scalability, flexibility and affordability

Winthrop University was certain that there had to be a solution that could deliver enterprise-class storage functionality with the scalability, performance and price flexibility they needed. After researching their options, the college discovered Gridstore.

“Once we began to look for new storage solutions to upgrade our SAN and NAS environment, we heard about Gridstore,” says Hammond. “We called Gridstore and scheduled a product demonstration. Based on what we saw and the price point to get started, we opted to skip test driving demo units within our environment and placed an order right on the spot. At a third of the price we paid for our SAN solution, the Grid offered a value proposition we couldn’t pass up. With great scalability, superior cost control and automatic optimization for our applications, there was no need to overprovision our storage as we had to before.”

The Grid uses patented Server-side Virtual Controller™ (SVC) technology to deliver automatically optimized performance for all applications across both virtual and physical servers. By combining simple building blocks of storage with intelligent vControllers, the Grid grows more powerful as more storage nodes and vControllers are added, allowing Winthrop to independently scale capacity, IO and processing, at a fraction of the cost of traditional storage offerings.

Winthrop University deployed the Grid to support its disk-to-disk backup processes. Since the school had reached the limits of its DAS storage, it needed a quick, cost-effective solution to upgrade storage in support of its data protection strategy.

Based on its positive experience with the Grid in its back up infrastructure, the university will leverage Gridstore to optimise its Hyper-V environment, with a goal to achieve 100% data redundancy for its virtual infrastructure in an off-site data centre. Winthrop has selected the Grid to help it achieve this goal with a storage solution that grows more powerful as it scales while eliminating IO performance bottlenecks, coupled with the value of a “Pay-as-you-Scale” flexible pricing model to achieve both cost and performance efficiency.

“We are already seeing Gridstore deliver a superior storage solution for more than one third the cost compared to our SAN, maybe even better,” says Hammond. “It has both fit my economic needs and future growth plans as we move to a virtual Hyper-V environment. Gridstore is an ideal storage solution for our IT infrastructure, both today and tomorrow.”

Gridstore and the Channel

Gridstore has designed its Accelerate Partner Program to ensure their partners can differentiate themselves from the competition and deliver the innovative storage solutions that todays’ customers demand.

“Our value proposition for the partners in the Accelerate Program is to deliver innovative storage solutions that aid them in maintaining their competitive advantage,” says John Ferraro, Vice President Worldwide Channel & Sales, Gridstore. “Our high-performance optimised storage provides the ideal cost-effective solution. It provides the perfect opportunity for channel partners to leverage their virtualisation expertise and ride the wave of the increasing penetration of Hyper-V.”

As a 100% channel-driven company, Gridstore sells their products exclusively through a highly selective network of value-added resellers and service providers that have the virtualization, Windows and technical skills necessary to deliver IT solutions for the today’s demanding data centre.

Ferraro continues: “Our Accelerate Partner Program has been designed to offer our partners the assets they need to be successful. This includes attractive margins, live technical training, marketing support and tools, deal registration, pre-sales support, highly responsive deal desk, and an informative partner portal. Together our solutions and the Accelerate Program will enable our partners to profitably expand their business and deliver true customer value.”

The dramatic growth in data over the past few years has led to an exponential increase in the number of storage solutions designed to deal with the rise. As a result it has become increasingly difficult for those in the channel to stand out and have an edge over the competition. Gridstore’s partner program is designed to provide the tools, resources and experience to ensure that partners are fully equipped with the building blocks to successfully promote the solution.

“We are expecting 2014 to be the year SDS comes of age. It will move away from being a nebulous buzzword, becoming a reality for a great number of customers and we are keen to ensure our partners are at the forefront of the opportunity this presents. Our SDS solutions allow partners to provide virtualisation offerings that are easy-to-use, easy-to-manage and enable simple scaling while reducing IO bottlenecks,” states Andy Hill, Vice President of EMEA Sales, Gridstore.

Hill adds: “Finding and working with great partners is core to our UK and European growth strategy. We are going to need to leverage their support and local expertise to help us hit our ambitious growth targets.

“In return they will receive a genuinely innovative and market leading product. We believe that this is an exciting time to partner with us as we introduce our solutions to a new market that is more than ready to take them on board. The market is worth more than £11bn a year providing plenty of opportunity for the channel to be successful. The key is to stand out from the crowd. Our SDS solution combined with the ability for partners to offer pay-as-you-grow customer acquisition model has and will continue to prove a great business model and one that can fit seamlessly in the UK market.”

Box

S3 and Gridstore – DCS talks to
Mark Smith, S3 Technical Director

Q What made S3 decide to become
involved as a Gridstore value-added
reseller?

A S3 are always looking for innovative
products that address growing
markets. We have seen a shift
towards Hyper-V in some of our
verticals and there seems to be
more drive to reduce the cost of
the test and development
environments – hyper-v is an obvious
choice and combined with Gridstore
makes a compelling proposition.

Q What does the Gridstore technology
add to the S3 portfolio?

A Scale-out rather than Scale-up is a
hot topic. The I/O blender effect
that some traditional storage
approaches just doesn’t work for all
environments. The Gridstore
architecture enables pay as you grow
with some interesting technical
features.

Q What are the benefits to S3 of working
with Gridstore – ie how does
Gridstore support its Channel
partners?

A S3 always work
closely with our
vendors, the
closer the better. With the direct
purchase model that Gridstore
employs, we are directly engaged
with the Gridstore team and equally
invested in the opportunities and
results.

Q What success has S3 had with the
Gridstore technology to date?

A Gridstore is a new vendor for S3 and
we are still ramping up sales and
marketing effort. The feedback from
our customers has been very positive
and we are looking to convert three
proof of concept projects by the end
of this quarter alone.

Q What is the Gridstore sales pipeline
looking like for 2014?

A It’s looking good. New products
can be either hot potatoes or slow
burners, in this instance we see
Gridstore as a technology that is
easy to understand and delivers
some compelling benefits so the
pipeline is building nicely.

Box 2

Gridstore’s Kelly Murphy on:

Windows Hyper-V

Gridstore has made a big bet on the adoption of Hyper-V. This bet is beginning to pay off. According to the latest number from IDC, as of Q3 2013, Hyper-V had 31.5% of the market and was growing. The product has a lot going for it. Not only is it low cost (almost free for those with enterprise Windows licenses), but it is also very solid and has the functionality to meet most customer needs. Plus it keeps getting better. With the release of the gen2 Windows virtual machines, we expect our customers will see significant improvement in application performance over what they see today on either VMware or Hyper-V, but this will only work with Hyper-V.

A recent survey that we conducted shows similar numbers that IDC saw, and if I look at our core market, the mid-market enterprise customers, I think the percentage is even higher.

Furthermore, I think that Europe has a lot more penetration of Hyper-V than the US – as a direct result of virtualisation being adopted somewhat later.

Hybrid storage

Flash is clearly a major game changer for the storage industry. I believe most of the all flash solutions are still addressing a niche that requires 1 million or more IOPS. They try to justify their prices with primary storage inline deduplication technology. Dedupe is a function that will be everywhere over time, so that strategy doesn’t hold up over time. The price of flash will continue to come down, but we are a long way from pure flash systems become mainline. And let’s not even talk about the death of spinning disks, we have had that discussion about tape for the last 15 years.

Hybrid flash has a lot of advantages. Deduplication is not necessary to offer a very cost effective high performance option. Even though you don’t normally see 1 million IOPS, you do see 20,000 to a few hundred thousand IOPS from hybrid solutions. This is the sweet spot for most applications that require high performance. I expect to see hybrid flash in greater and greater numbers this year.

Application performance

We conducted a survey in the summer of 2013 about different aspects of virtual environments. When asked what the respondents top challenges were in virtual environments were, I was not surprised to hear that number one was getting the necessary application performance. Once an application is virtualised, it is competing for resources. Often the top limiting resource has to do with the storage stack.

There is the problem of the “I/O blender effect”, the limitations of a LUN serving many different applications and having lower priority applications hogging network and controller resources. I was a little surprised to see the number two challenge was the cost of the storage hardware for these environments. After thinking about it, the answer was obvious. For those who tried to solve the application performance issue with over provisioning, that made the cost of the storage so high. That can’t continue. This is why QoS is becoming very important in virtual environments. I expect QoS will be as important as deduplication over the next 18 – 24 months. Clearly vendors that provide a cost effective, good performing storage solution that scales to meet the needs of the virtual environment and delivers QoS to keep the noisy secondary applications quieter, will see success.

Scalability

Scalability is obviously important. Data continues to grow at exceptional rates. Unfortunately budgets do not. With more public and private clouds, organisations need to react to the growing needs for storage, but without purchasing far ahead of the needs. You want to add storage as the organisation needs it. You want to pay for this as the organization needs it. But you don’t want to have to hit a point where you are going to have to rip out what you have purchased so far and replace it because it can’t keep up with the scale requirements. This will drive the desire for scale-out solutions that are not limited by clustering or the requirement to purchase more than you need at any time.