Experts disagree during sessions at Storage Networking World
ORLANDO — Grid storage and storage clusters — are the terms interchangeable? They could be if you ask users and industry analysts who participated in several panels at Storage Networking World this week.
There was a lot of confusion at this show as to what makes up a storage grid and how it differs from a cluster.
“If I think of the two or three technologies that are known for the greatest amount of conjecture, disagreement and confusion, grids and grid storage would be right up there,” said Simon Robinson, senior analyst for the 451 Group and moderator of the panel on grid infrastructure.
Paul Strong, a distinguished research scientist, has a storage grid installed at eBay in San Jose that consists of two petabytes of capacity. His grid grows by 75 data volumes and 10TB of storage per week.
“At eBay, we consider our grid to be the underlying storage fabric,” said Strong. “The issues we have with the grid is its ability to scale and agility to change. Management of the grid is complicated and hard. We are looking to move away from managing bits, bytes and blocks of data, and deploying software, technology and standards that can manage the grid as a single entity.”
To do so, Strong called for standards and software for managing storage grids. Matthew Brisse, technology strategist in the Office of the CTO at Dell, who presented on grid architectures at the show, agreed.
“We’re not there yet,” said Brisse, adding that much of the middleware software that manages the grid is not available.
|
“Grid-based storage is storage that has grid attributes associated with it, meaning an orchestration middleware layer with its own quality of service capabilities, a scheduling layer that distributes the tasks on the grid, and a set of business policies that manage the cluster of computers and storage,” he said.
Brisse is working with the Storage Networking Industry Association, one of the sponsors of Storage Networking World, which launched a Grid Task Force last year. The objective of the task force is to work with the Open Grid Forum to develop a grid proposal and standards that will help companies deploy grids.
Are grids then distinguished from clusters simply by the addition of middleware, scheduling and policy-based software that organizes, controls and manages the grid and its operation?
A storage cluster, Brisse said, is a network of multiple storage systems that behave as a single, tightly coupled system. A grid links disparate pools of distributed storage and server resources together with services and software in a loosely coupled structure that can be managed as a utility.
“Is it possible to have a single storage array that has compute nodes that can scale [in performance or capacity] like an EMC Centera or a Permabit cluster — is that a grid?” said Brisse. “Technically, no — grids have to be geographically dispersed.”
John Hurley, head senior manager of distributed software and systems integration at Boeing in Seattle, has taken Brisse’s definition of a geographically dispersed grid to heart.
“Many of us have become global companies, and that’s changed tremendously how we look at data,” said Hurley, whose grid spans from Puget Island, Wash., to St. Louis, to Madrid.
“We typically implement policies and practices for data and how users access that data,” said Hurley. “The basic misconception about grids is worrying about the definition. I don’t care about that. I need my problem solved.”
Although Hurley would not discuss the physical infrastructure of Boeing’s grid or its financial investment, the grid at Boeing is seven years old and originally deployed to cut costs and share resources among the company’s geographically dispersed sites. Alan Yoder, senior technical staff at Network Appliance and co-chair of the Open Grid Forum, would agree with that assessment.
“A storage grid is a utility that is basically a bunch of storage that is controlled by some central entity that virtualizes it and makes it into a utility,” said Yoder. “You treat data as a service, storage as a service, all of which are universally accessible from within the grid.”
However, Greg Schulz, senior analyst for StorageIO, said many users confuse grids with clusters and that in the end, there is little difference.
“There’s no difference between a grid and a cluster,” said Schulz. “If you look at an OnStor, Lefthand, Equallogic or Isilon clusters, those are all grids. They’re all clusters too. A grid is a cluster, a cluster is a grid. A grid is a product, a technology, an architecture and a service. The other way of looking at a grid is as a utility. If you can’t use the utility model, call it a grid service.”
