Server-Side Virtual Controller™ Technology (SVCT)

All Traditional Storage Suffers from the I/O Blender

All traditional storage suffers from the I/O blender that severely impacts application performance in virtual environments.

Prior to virtualization, an application server OS would optimize the I/O specifically for that application before moving the I/O efficiently to the storage subsystem.

When a server is virtualized, the OS still optimizes the I/O, but the I/O produced is then blended with 10, 20 or even 30 other virtual machines running in the same environment.

By the time the I/O reaches the storage controller, all the I/O has been blended together from possibly hundreds of virtual servers and there are conflicting I/O patterns that result in as much as a 10X performance loss, even if the I/O is optimized.

Even the latest all-SSD storage array or hybrid array continues to use the same inefficient architecture. And while they improve performance, they are an expensive, brute force approach that does not address the real problem. High performance arrays still cannot optimize or prioritize for each different workload and they force organizations to pay a premium even for those applications that do not need such high performance.

I/O blender impacts both virtual and physical servers

Just as the virtual environment is severely impacted by the I/O blender, if your physical servers share the same storage system, they are also infected with the performance problem.

And if the physical servers are affected, then the primary workloads running on those physical servers can experience that same performance hit due to strained storage controllers trying to deal with the random I/O coming from possibly hundreds of virtualized servers.

Gridstore Eliminates the I/O Blender

Gridstore is Software-Defined Storage. It uses patented Server-side Virtual Controller Technology (SVCT) to eliminate the I/O blender at the source, align storage to the virtual server model and accelerate application I/O for every VM.

SVCT virtualizes the traditional storage controller and moves it across the network to the application server. By operating in the application server, the virtual storage controller—or vController™ creates a virtual storage stack for each that isolates, optimizes and prioritizes I/O for each VM before it leaves the server.

vController™ — Requires no change to Hypervisor or Guest VMs

A vController is a native Windows driver certified by Microsoft. It presents a local SCSI block device to the server. The device is instantly loaded without rebooting the server. The server finds a new disk that can be instantly formatted with NTFS or CSV2 and made available for use within minutes. The disk is managed the same way you manage any diskin Windows environment. vController software can be pushed to any number of servers through standard Microsoft management tools.

vmOptimized™ Storage — Accelerates I/O for every application

vControllers isolate the I/O from each VM and create a virtual storage stack for each VM. The vController architecture then extends these channels across the storage fabric – creating virtual isolation lanes from VMs to storage. This completely eliminates the I/O blender before it gets started.

With the I/O isolated into separate channels, vmOptimized can dynamically optimize the I/O according to the needs of the application. Every application has an optimal I/O pattern – whether it is small random I/O or large sequential writes or anything in between these. vmOptimized will detect the I/O signature of the application and optimize it. Now VM’s with completely conflicting I/O patterns will not impact each other and each will be optimized for what they need.

vmOptimized accelerates every application without the need for manual tuning. vControllers with vmOptimized aligns the storage model directly with the virtualized server model and optimizes on a 1:1 relationship.

TrueQoS™ — Guarantees performance for your most important applications

SVCT creates the industry’s first end-to-end storage QoS – TrueQoS. By creating separate I/O channels for each VM, the vController can allocate resources such as memory, queue depth, thread pools, cache, IOPS and throughput to each vChannel based on policy set by the hypervisor. This policy is enforced across all hypervisor’s and physical servers and from VM to virtual storage pools.

SVCT creates an intelligent grid fabric to achieve this end-to-end capability. By distributing intelligence to the end points—both server-side and storage-side, SVCT can orchestrate I/O from end-to-end. Each end point in the grid is intelligent and coordinates to optimize and prioritize the flow of I/O between VMs and storage.

Separate I/O channels from the VM of the I/O through to the Storage Nodes. By distributing intelligence to both ends, SVCT controls the I/O flow end to end for all hypervisors and physical servers to create.

TrueQoS is the industry’s first end-to-end storage QoS and guarantees performance for your most important applications.

GridScale™ – Massively parallel performance scaling

When I/O leaves the server, the vController first protects it and then stripes it across parallel network connections to multiple storage nodes in the vPool. Each I/O is broken into equal parts as storage nodes that make up the stripe in the vPool. These parts are written directly and in parallel to storage nodes – we call this Direct I/O.

Direct I/O eliminates the need for a 3-way replica (common to scale-out storage models), complex clustering, load balancing and a backplane network to carry 3X the I/O. Direct I/O saves on both CAPEX and OPEX, while the elimination of the clustering overhead delivers 3X IOPS with significantly lower latency.

With GridScale – there is no inter-node traffic – each storage node does a fraction of the work. The more storage nodes there are, the less work each one does. This allows the kind of scaling of capacity and performance to levels normally on available in High Performance Computing environments or what has become known as Hyperscale environments on the web.

By breaking the barriers of clustering, GridScale delivers the massively parallel performance of up to 250 H-Class nodes per vPool™ — equal to 3PB capacity, 1000 Xeon cores, 8 TB RAM, 5 TB bandwidth and 200 TB of PCIe write-back caching.

GridScale gives you the ability to start small and seamlessly grow to a Hyperscale environment and pay-as-you-scale.

OPTIMIZED STORAGE FOR WINDOWS SERVERS AND HYPER-V