Comment In a storage cloud, empty bit buckets on disk platters are like a centrifugal force spinning money off and away. They are destroyers of value. Unless storage suppliers can deal with this, they're going to be outclassed and uncompetitive.
David Scott, 3PAR's CEO, says his company is set to bring new features to its InServ arrays that will benefit cloud service providers.
Why is 3PAR looking to develop its product along these lines, making it more attractive to service providers? Scott said: "Most enterprise IT will be delivered as a service by 2030." That means 3PAR has to tailor its products for, and sell to, service providers, not the enterprise end-users who buy a lot of its kit today. The whole enterprise storage industry is involved in a multi-year transition to enterprise clouds and on to clouds for enterprises. That means the storage buying power is increasingly passing from enterprise end-users to the service providers.
Coming new features will enable them to benefit more from thin provisioning, for example, so that they stay thin over time and don't find storage provisioning fattening up through empty storage accretion. Filesystems, for example, should tell a block storage device when a file is deleted so that its allocated space can be recovered and re-used.
The storage should manage itself more - be autonomic - in the coming next-generation virtual data centres. There should be more resiliency, with multiple write-through streams and better long-distance disaster recovery. The storage platform should use solid state drives (SSDs) to optimise cost/GB and cost/IOPS with data being moved automatically between SSD and HDD tiers of storage as customer circumstances, data type and activity demand. The platform should enable differentiated quality of service for different applications. (This sounds like Pillar Data speaking, as its Axiom arrays do just that already.)
This is going to involve fine-grained control of storage space. Scott contrasts 3PAR's ability to operate from 256MB chunklets down to 16KB blocks with the coarse-grained control in EMC's Symmetrix V-Max. This has automated LUN movement between tiers coming and then sub-LUN movement following many months behind that. He says Symmetrix has no autonomic policy-management layer either.
All these things need to be available for different sets of applications for the many different tenants to be expected in a cloud service environment. There are hints that some or all of these could come from 3PAR in the next month or so.
Scale and reduplication
Cloud service providers need storage scale as well. Scott reckons 3PAR has a great advantage with its proprietary ASIC, high-speed internal networking and clustering, contrasting this with NetApp and its ONTAP 8 clustering O/S release via enabling multiple heads: "You cannot scale a block storage application without a low-latency interconnect."
Sheer clustering scalability, adding more nodes, isn't desirable: "3PAR could double or triple the cluster count of 8 nodes, but a software failure could have catastrophic effects with a 64-node cluster coming to a halt. It's better to federate and provide manageability. You have to understand the failure domain. There is a formula combining the risk perspective and an application workload which determines the optimum cluster node count. You don't go beyond this."
Scott is doubtful about the use of deduplication in primary storage in the cloud service scenario or in general: "If you apply this in the primary storage area, you start impacting response time. You theoretically could provide hardware-based compression but it might not work well with a thin environment." Thin provisioning and fat-to-thin migration strip out the repeated zeroes found in over-allocated space, thus removing one of the main targets for compression.
The features needed are not glamorous or sexy, not eye-catching like Dataram's idea of a flash cache between Fibre Channel switches and the block arrays connected to them. But they mean a lot to 3PAR customers like Savvis, because it enables them to sweat their storage assets, to wring cost out of them.
When part of your core business is renting out storage space, wasted bits and costly management equate to lost revenue and profit. Small instances of these can be tolerated in one or two data centres owned by an end-user enterprise, but not in the global data centre estates operated by cloud service providers. There, small instances multiply into huge ones, and empty bit buckets are holes through which dollar bills, euros and pound notes float away. ®