Hitachi Data Systems (HDS) has its own take on EMC's 1 petabyte (PB) plus DMX-3 announcement. It is that EMC did what it could and what it could do was limited by the Symmetrix architecture. Whereas EMC can have a 1 petabyte DMX-3 array HDS' Tagmastore USP (Universal Storage Platform) can only have 345TB (0.345PB) of internal storage. But TagmaStore USP can manage 32PB of storage in total by using external (and heterogeneous) arrays and presenting it as a single 32PB of logical storage. That's at least thirty times more.

EMC can't do this. For HDS it's all to do with its virtualisation facility, its Tiered Storage Manager (TSM), its multiple storage link protocol support and, crucially, its ability to virtualise external disk arrays.

I asked Roger Turner, HDS director of hardware product marketing, about this difference between the EMC and HDS approaches to high-end storage.

TW: HP and EMC use both Fibre Channel (FC) and Fibre Channel-attached ATA (FATA) drives in their arrays. What drives can you have inside TagmaStore?
Roger Turner: We only use proper, full duty cycle, Fibre Channel drives. TagmaStore is inherently tier 1 storage. For archival-class ATA drives we'd use external storage, such as our WMS100, and virtualise it. That way is less expensive for customers. Our approach simplifies the way you manage this.

TW:Can you have different tiers inside TagmaStore.
Roger Turner: Of course. You can optimise the storage and its quality of service to match server and application needs. You could have 73GB, high rpm drives for applications needing a low head to drive contention ration. There could be mid-capacity 146GB drives and high-capacity 300GB drives with a lower spindle speed. These could be used for applications satisfied with a high head to drive contention ratio.

TW: These would be less mission-critical applications. How are these different tiers organised and how is data moved between them?
Roger Turner: With TSM - our Tiered Storage Manager, which is a software toolset. It effectively manages the movement of data between tiers based on policies (set by customers) or on usage patterns. Servers don't need to know that the data they are accessing has been moved because its world-wide name remains the same.

TW: OK. You can have multiple tiers of storage and you can have one logical pool. Doesn't each tier form its own logical pool? How is this worked out?
Roger Turner: TagmaStore does present one logical pool but it is divided into logical partitions, roughly equivalent to mainframe LPARs. Each logical pool is a virtual machine in its own right with its own cache, front and backend. TagfmaStore can have up to 16 virtual machines and the NSC55 can have up to eight.

TW: So there is a three-level hierarchy here: a single logical pool at the top; divided into logical partitions; each of which can have multiple disk storage tiers. Why did EMC choose the approach it did?
Roger Turner: It's the best they could do given their architecture. Because they can't virtualise external storage they have to add extra tiers inside Symmetrix.

TW: Okay, grant that, but won't InVista be able to virtualise external, and heterogeneous, disk storage?
Roger Turner: You use the future imperfect tense! The big problem is that InVista can only virtualise FC-attached disk. InVista can't virtualise FICON or ESCON-connected arrays, nor can it virtualise NAS or SATA drives. So EMC put ATA (meaning FATA) into the Symmetrix. That's expensive. We keep it external and put it on lower cost modular ATA storage array.