I participated in GestaltIT‘s TechFieldDay which is a sort of inverse conference, where the bloggers and independent analysts go to the vendors and then discuss the information they have received. We visited the following storage vendors:
- Data Robotics where we were introduced to the new Drobo FS
- EMC where we discussed stretched storage and other interesting futures
- HP where we were introduced to the IBRIX products
Of the five vendors we visited, 3 were storage, and one was heavily involved with storage companies (Cisco). I have written about Drobo FS before and there is not much new to Drobo not already discussed.
What is interesting is that all these Storage Companies are looking at Storage quite a bit differently, no longer the monolithic storage array but the integrated array that may stretch over miles. The Drobo FS with its Oxygen Cloud integration, EMC with its thoughts on a stretched storage but with no real products yet, and HPs IBRIX solution which cannot yet stretch over long distances but is being considered.
So what do these products imply to the virtualization and cloud community?
Data Robotics — Drobo FS
Drobo FS, is the home users introduction to cloud, which is also what you get with the EMC StorCenter ix2 devices which includes Mozy integration. Granted the Drobo FS’ options seem a bit richer, with the possibility of bi-directional replication, but at the moment not available. The news from Data Robotics is that the Drobo FS is for the SoHo or small business as such it would not necessarily work as a data store for virtualization but an adjunct to existing replication that could happen from within a VM. The Beyond RAID technology, comprised of their micro-raid-set technology,is the meat of the Drobo products and allows the system to make use of different sized commodity drives to provide mirroring and striping. Listen to Drobo’s Tom Buiocchi talk to the GestaltIT Tech Field Day delegates.
HP’s IBRIX is really geared towards general purpose storage that needs automatic dispersal of data amongst multiple tiers of storage. It is quite possible to put one of the IBRIX file server gateways in front of existing storage arrays even from multiple vendors to join them to the single virtual array. IBRIX presents as a file system that spans multiple devices. The main HP selling point is that IBRIX is made up using industry standard components based on the HP DL380 and is comprised of compute and storage nodes. The compute nodes are used to do the heavy lifting of ensuring the data on the file system lives on the tiers of disk assigned to it based on performance and capacity requirements. IBRIX looks more like a solution for High Performance Technical Computing than it does for the virtual environment as most of the solutions spoken of were geared towards many bits and pieces of discrete data or small reads and writes over larger volumes. The IBRIX does automatic tiering of data based on access times, but leaves the access path intact. Where the data was written within the IBRIX file system may change but how it is presented to the applications using the data does not. To use IBRIX auto-tiering with virtualization such as vSphere would require the auto-tier rules to understand what comprises a virtual machine as they are not single files.
HP IBRIX provides at the file level what Lefthand iSCSI provides at the block level using similar hardware and software. While Lefthand is an iSCSI SAN, IBRIX provides high speed NFS capabilities.
EMC discussed its vision for a stretched multi-geographic storage cloud that would allow for fast movement of data from one location to another. The theoretical discussion was about using FAST and pre-fetch capabilities to perform long distance data migrations. The FAST aspect would possibly automatically tier data simultaneously between multiple locations. They followed that theoretical discussion up with concrete discussions on the three aspects of storage requirements, capacity, and performance. Nearly all of the items mentioned are well understood from multi-tiered storage, cache, modularity, etc. However the main ingredient not well understood is VMware Integration and Hardware Offload capabilities. The vStorage API for Array Integration (vAAI) offers, when it is available, quite an improvement in overall storage performance with respect to virtualization. Many of the repetitive tasks would be pushed down into the array, just like Intel-VT and AMD-V pushed many of the CPU intensive tasks into the CPU, vAAI has the chance to do the same for storage.
EMC’s vision includes tight integration between storage and virtualization management tools such as vCenter which will provide the flexibility required. The vAAI component (performance) of this vision is actually the more interesting of the two points. Within a virtual and cloud environment storage is a major concern. However, if EMC is the only one supporting vAAI, within a VMware environment they will have a huge competitive edge. I expect that it will not be long before the other storage vendors either develop their own or also program their arrays to support vAAI, as this is truly an array issue. If the hypervisor can detect vAAI is available it would be foolish not to use it.
One of the first vAAI aspects to be worked on is the zeroing of disk blocks. Currently when vSphere builds a disk or expands a disk on the fly due to thin provisioning within the hypervisor, vSphere sends down to the storage array a large number of zero that fill the allocated blocks of disk. Just offloading this one aspect will leave more cycles within the hypervisor for other actions and speed up the hypervisor’s version of thin provisioning. Listen to Chad Sakac of EMC talk to the GestaltIT Tech Field Day delegates.
Storage vendors are aiming for the cloud using at least three different styles of entering the cloud. Drobo FS wants to be an end point, HP IBRIX wants to be the back end, and EMC wants to be integrated into the virtualization and cloud management stacks while providing the end point and back end storage.
EMC provided a very consistent view of storage for the future with vAAI and management console integration. The EMC Celerra NFS plugin to vCenter is a case in point. EMC is thinking quite hard about how storage management will take place in the future. The fact that virtualization administrators may need greater access to the storage devices. With vAAI, EMC is working to increase the overall performance within the virtual and cloud environments.
HP does not. HP has to marry various existing storage technologies into a consistent view. HP mentioned IBRIX, Lefthand, EVA, and Tape devices as they move a ton of tape. HP’s traditional SANs (EVA) are apparently not dead. But we do not have a consistent view of how they fit into the
Data Robotics on the other hand is in a position to own the prosumer space as an entry into the cloud, but they have teamed up with a cloud provider not many people know very well.
For vAAI to be a success and not provide EMC a distinct competitive advantage, other storage vendors need to be involved.