A Practical Scale-out Storage Solution minus Scale-out's Costs and Complexities
Server virtualization is creating new demands for ease and simplicity in managing and scaling storage capacity and performance that go well beyond what traditional single or dual controller storage systems can provide. But what organizations may fail to recognize is that even current scale-out storage systems may NOT be truly optimized to cost-effectively deliver the storage capacity and performance that virtualized environments require. This is the unique value proposition that Gridstore with its vController technology is well positioned to address.
Contemporary scale-out storage solutions overcome some of the problems of traditional single or dual controller storage systems by clustering the nodes together and tightly integrating a private backplane network with storage nodes and controllers. This integration gives organizations more flexibility to linearly scale-out capacity and performance in a modular fashion while easing storage management. For example, this technique minimizes or even eliminates data migration challenges as data may be seamlessly moved between nodes to improve performance or facilitate high availability.
Yet scale-out storage systems have their own cost and complexity issues. A few take the approach of adding a clustering feature to their legacy single or dual controller systems with an upgrade to their existing controller software. The end result here is a storage configuration that these systems were never originally designed to provide or support.
Most designs rely upon a clustered set of nodes to improve redundancy and performance by sharing the load among a group of storage nodes. However to provide both a unified storage and management interface to clients and administrators, all data must pass through powerful storage controllers that can be both costly to acquire and complex to configure and manage.
Regardless of which of these two approaches is adopted, they both dictate that all storage processing occur on the controller. This forces these solutions to use high performance hardware as storage controllers to host sophisticated software that handles and directs the storage networking traffic.
While enterprise organizations may be in a position to absorb these controller costs and manage their complexity, midsized companies typically are not. However these smaller organizations still desire the flexibility that such a scale-out architecture provides.
So to deliver these scale-out storage benefits small and midsized organizations want without the cost or complexity, Gridstore has moved beyond the requirement for physical controller nodes to provide a virtual grid storage solution.
In Gridstore's configuration the controller is virtualized and moved from the storage node to the client accessing the node. This approach moves the data processing out to the client and eliminates costly controllers. The strength of this approach is that these virtual controllers, or vControllers, are implemented without any extra cost. Further, moving the workload out to the client provides several advantages over the traditional scale-out storage model.
The vController's combination of parallel I/O and distributed data processing also means that replication and recovery performance of storage nodes is enhanced. The initial work of identifying what data needs to be distributed between nodes is done by the clients in the grid. Each client then writes its data in a point-to-point fashion in parallel to several storage nodes with parity data to ensure fault tolerance. Written in this way if a single node should fail during the write the data will not be lost. This technique ensures both higher levels of fault tolerance and faster recoverability as nodes leave or join the grid.
This distributed approach to accessing data may also enhance application performance. As more storage nodes are added, the data is spread out among more nodes to allow more clients to access it at the same time. In the case of Gridstore, this data distribution happens on the fly with no interaction necessary by a user, easing administration even as it increases performance.
The plug and play aspect of Gridstore coupled with the fact that it runs on simple commodity hardware provides a tantalizing solution for small to medium sized companies looking for storage solution that can fulfill several roles within them.
Instead, using its zero-cost vControllers that reside on clients accessing its storage nodes, organizations can improve their storage availability and increase application performance even as they drive down their storage costs. Further, Gridstore addresses key complexity concerns that small and midsized organizations possess as its storage nodes are easy to manage since they are largely plug-and-play.
But what is possibly most interesting about the Gridstore solution is that it does more than enable small and midsized companies to cost-effectively implement a scale-out storage solution for their archiving and file serving needs. It positions them to confidently use a scale-out storage solution in their emerging virtualized environment where performance is potentially much more demanding and unpredictable than in today's physical world.
Contemporary scale-out storage solutions overcome some of the problems of traditional single or dual controller storage systems by clustering the nodes together and tightly integrating a private backplane network with storage nodes and controllers. This integration gives organizations more flexibility to linearly scale-out capacity and performance in a modular fashion while easing storage management. For example, this technique minimizes or even eliminates data migration challenges as data may be seamlessly moved between nodes to improve performance or facilitate high availability.
Yet scale-out storage systems have their own cost and complexity issues. A few take the approach of adding a clustering feature to their legacy single or dual controller systems with an upgrade to their existing controller software. The end result here is a storage configuration that these systems were never originally designed to provide or support.
Most designs rely upon a clustered set of nodes to improve redundancy and performance by sharing the load among a group of storage nodes. However to provide both a unified storage and management interface to clients and administrators, all data must pass through powerful storage controllers that can be both costly to acquire and complex to configure and manage.
Regardless of which of these two approaches is adopted, they both dictate that all storage processing occur on the controller. This forces these solutions to use high performance hardware as storage controllers to host sophisticated software that handles and directs the storage networking traffic.
While enterprise organizations may be in a position to absorb these controller costs and manage their complexity, midsized companies typically are not. However these smaller organizations still desire the flexibility that such a scale-out architecture provides.
So to deliver these scale-out storage benefits small and midsized organizations want without the cost or complexity, Gridstore has moved beyond the requirement for physical controller nodes to provide a virtual grid storage solution.
In Gridstore's configuration the controller is virtualized and moved from the storage node to the client accessing the node. This approach moves the data processing out to the client and eliminates costly controllers. The strength of this approach is that these virtual controllers, or vControllers, are implemented without any extra cost. Further, moving the workload out to the client provides several advantages over the traditional scale-out storage model.
- All data is broken into blocks and striped across the storage nodes in the volume
- A list of all nodes in the grid is maintained on each client allowing direct, point-to-point communication with storage nodes
- During the discovery period each client communicates directly with the storage nodes to determine which nodes contain the data blocks it needs
- Once the client discovers which storage nodes host the data it needs, it does not need to interaction with other storage nodes freeing them to directly service other client requests
- Point-to-point communication provides true parallel IO throughout the grid which improves the grid's overall throughput
- Performance is further improved by the vController optimizing data requests before they leave the host as this eliminates duplicate requests.
The vController's combination of parallel I/O and distributed data processing also means that replication and recovery performance of storage nodes is enhanced. The initial work of identifying what data needs to be distributed between nodes is done by the clients in the grid. Each client then writes its data in a point-to-point fashion in parallel to several storage nodes with parity data to ensure fault tolerance. Written in this way if a single node should fail during the write the data will not be lost. This technique ensures both higher levels of fault tolerance and faster recoverability as nodes leave or join the grid.
This distributed approach to accessing data may also enhance application performance. As more storage nodes are added, the data is spread out among more nodes to allow more clients to access it at the same time. In the case of Gridstore, this data distribution happens on the fly with no interaction necessary by a user, easing administration even as it increases performance.
The plug and play aspect of Gridstore coupled with the fact that it runs on simple commodity hardware provides a tantalizing solution for small to medium sized companies looking for storage solution that can fulfill several roles within them.
- Archiving. Gridstore's innate ability to provide fault tolerance makes archiving a natural use case. Adding storage capacity is as easy as plugging in an additional storage node into the Gridstore solution plus it increases its fault tolerance.
- File server. Gridstore's scale-out capability combined with the performance advantages of its grid model makes it a natural fit as a file server.
- Virtual Desktop Infrastructures (VDI). The parallel IO aspect of the Gridstore model lends itself to alleviating performance bottlenecks that occur during times of peak demand, such as when users arrive to work and all log in during a short period of time.
Instead, using its zero-cost vControllers that reside on clients accessing its storage nodes, organizations can improve their storage availability and increase application performance even as they drive down their storage costs. Further, Gridstore addresses key complexity concerns that small and midsized organizations possess as its storage nodes are easy to manage since they are largely plug-and-play.
But what is possibly most interesting about the Gridstore solution is that it does more than enable small and midsized companies to cost-effectively implement a scale-out storage solution for their archiving and file serving needs. It positions them to confidently use a scale-out storage solution in their emerging virtualized environment where performance is potentially much more demanding and unpredictable than in today's physical world.
Leave a comment