The storage array controller has primacy
HDS' storage vision
By Chris Mellor | Published: 19:00, 21 April 2005
HDS has set out on a distinctively individual storage product path. Sun, an OEM and possibly strategic partner, is ploughing a similar furrow. This involves the placing of storage service management intelligence at the front of disk arrays. It's not in the accessing servers; nor in appliances attached to fabric switches and directors; nor in the switches and directors themselves.
Instead it is in what we can loosely call very intelligent disk controllers, what IDC might describe as network storage processors. The fabric is relegated (my term) to a data-moving pipe, not a storage service-managing entity. Thus HDS' TagmaStore, a monolithic enterprise disk array, provides multi-vendor virtualisation and storage services. It has had a NAS capability added too, to emphasise its primacy.
In this HDS is radically diverging from approaches espoused by virtually every other storage vendor, apart from Sun. (Some newer suppliers also place intelligence in the array controllers, like 3Par with its InServ array and services but the majority pof vendors locate their virtualisation and storage management platform either in the fabric or in servers.)
Techworld talked to Vincent Franceschini, HDS' senior director, future technologies, about HDS's view of storage and it's directions.
TW: What does virtualisation mean for HDS?
VF: We're trying to understand what virtualisation means for the end-user. The original point products saved customers' disk blocks and helped consolidate storage across vendors. We belive consolidation is key. It needs to scale; that's a very important aspect; from hundreds of GB to petabytes. Different users along this range share some common requirements.
But it's not just about capacity scaling, its also about performance, data protection and other features. It's not something that's isolated. Virtualisation is a foundation technology for other functions, such as DLM (Data Lifecycle Management). Virtualisation helps customers achieve more.
We're trying to manage the storage resource. We can virtualise other vendors' storage but without a management layer that's meaningless. That's why we did Storage Manager (from AppIQ).
TW: What about its placement?
VF: Putting virtualisation and management in the fabric makes it much more complex. Also (coming at it from the fabric point of view) means supppliers like Cisco have a network-oriented quality-of-service (QOS). We see a need for a data-oriented QOS.
If you put virtualisation and management in the fabric there is a scaling factor. For example, you can snapshot thousands of LUNs out of a TagmaStore. How are you going to do that in the fabric? How big does the switch have to be?
These two solution types may have to co-exist. There are different customer requirements.
TW: Tell us about DLM.
VF: It is really about making sure the storage infrastructure is in sync with higher level requirements. Via policy-based systems we can drive where data is located on the storage network and define the criteria to move data around. As we expand our (management software) suite we will integrate more application-aware parameters.
TW comment: We can envisage a storage tiering system like a matrix with axes based on performance, media cost, protection level, a quality of service level and a file or folder's metadata.
TW: What about tape and optical media?
VF: DLM embraces both tape and optical media. Tape's role now is as a long-term archive, being fed by the lowest tier of disk storage, where data is in transit. We will have an API structure in our Tiered Storage Manager that will allow us to engage with ISVs.
TW: Is HDS heading towards providing a storage utility?
VF: We're evaluating the notion of grid computing and what it means. One thing is that it's about managing data security. We might have to manage different levls of security within a volume.
A grid approach can help us to be very modular and very flexible with our future architecture.
We see some kind of merger happening between SAN and NAS requirements. We must view a storage service as a universal service for the customer. Many customers want us to integrate NAS and SAN.
TW: What about a global file system?
VF: A global file system (GFS) is a next step. With TagmaStore we have a very powerful and very flexible storage centre and can develop extra storage services on top of it, i.e. GFS or continuous data protection.
TW: Could you tell us more HDS's views on NAS apart from the TagmaStore level?
VF: We will consider a modular-based NAS solution. It's something we need to address. We started with the enterprise first because that's where the maximum pain point is.
(Compared to Network Appliance whose NAS filers HDS resells) We've developed a different NAS strategy. But it's still a quality relationship with Network Appliance and will carry on like that. But we see our needs in the storage industry going in a different direction.
TF: Will you provide virtualisation and storage management services, like the TagmaStore facilities, for the Thunder and Lightning arrays?
VF: We don't see any traction for it now. But it's not dismissed.
TF: IBM has its Power5 procesor and Sun has its Sparc chips. What is the CPU in TagmaStore?
VF: Our Intelligent hardware is the fruit of joint R&D from the HDS server, storage and communications divisions. The CPU debate is driven by the traditional server vendors. You don't just need a fast CPU. It's about the whole architecture, about managing control and data separately, about having a systemic approach.
TW: HDS is going to provide NAS for businesses and business needs below the enterprise level, using modular arrays. Yet HDS is also seeing a strong need for consolidation at this level too. How might it provide this?
VF: The mid-level enterprises will need consolidation too. We could drive TagmaStore downward, make it more affordable. It would enable existing storage arrays to be consolidated and provides a level of service that can't be achieved with a modular-only tactic.
TW: Does HDS believe that 2.5 inch disks are needed to increase the I/Os per second from an array or array shelf? (This is the Seagate concept.)
VF: Seagate would say that. It's trying to catch us up in the 2.5 inch market. We are the world leaders in 2.5 inch drives. For now we don't see that the idea of moving to 2.5 inch drives to increase the IOPS rate makes sense for the enterprise array.
Techworld comment: With the move to 4 Gb/sec Fibre CHannel and TagmaStore's internal communications architecture, Franceschini doesn't believe HDS' enterprise arays are IOPS-limited.
HDS has announced a serious and concerted push into the NAS market. Franceschini confirms that NAS products based on modular drive arrays, presumably Thunder and/or Lightning, are likely to come. But he also suggests that a down-sized TagmaStore might be in the offing, something that could provide a multi-vendor storage consolidation platform for the mid-tier enterprise market and something that could be a platform for virtualisation and storage services layered on top.
It will be interesting to compare this offering, if it comes, with Sun's 6920 which appears to be driven by the same ideas.