Young Lady or Old Lady- New Perceptions in Virtualization
May 23rd, 2008
This is an old advertisement for the anchor Buggy Company which shows a picture of a young lady and an old lady. I would guess that most of you would see the young lady and wonder where the old lady is. Most of you have probably seen the W. C. Hill version of this during your school days. Click on this link to see how to find the old lady.
http://mathworld.wolfram.com/YoungGirl-OldWomanIllusion.html
This is a clever illusion, an ambiguous drawing in which the brain will switch between different perceptions based on previous experience or external suggestions (marketing).
Some would say that storage virtualization is an illusion in that it can be perceived differently based on prior experience or marketing stimuli. Storage virtualization comes in many forms, from RAID groups, to virtual volumes, to virtual files, to virtual capacity or thin provisioning.
The virtualization of volumes was introduced nearly 25 years ago by Veritas Software with their Veritas Volume Manager. This was volume virtualization in the Server. Shortly after SAN’s were introduced many startups and many established companies tried to move volume virtualization into the SAN network. Their logic for placing volume virtualization in the SAN was that the SAN was a central place for connecting storage systems to servers and volume management could be consolidated out of the servers and onto the SAN. Suddenly everyone was in a race to recreate a volume manager layer in the SAN through the use of appliances or intelligent switch ports, and holy wars broke out between in band and out of band approaches to volume, or what they called storage, virtualization. Venture capitalists were eager to invest in the next big thing and several billion dollars were spent on this.
Many companies were successful in creating a SAN based volume virtualization systems. However, the volumes for this market did not materialize and many startups were sold off and many more went out of business. Unlike the IP network business, the SAN market is not a volume market in the first place. The main reason SAN based storage virtualization did not take off was that it did not fulfill the expectations for storage virtualization. While it did provide a degree of volume management, it did not mask complexity and it did not aggregate storage services to enhance lower level storage, and increase utilization. (See the SNIA definition for Storage Virtualization. http://www.snia.org/education/dictionary/s/ )
In the last seven to eight years the perception of storage virtualization has been that it must be done in the SAN since everyone is doing it there, including major vendors like EMC and IBM. There is also the perception that storage virtualization is too complex since it introduces another layer of hardware and management to map storage extents from physical volumes to virtual volumes and requires clustering to protect the mapping tables. It adds latency, especially if you are doing a copy or move of a volume, does not scale, and does not work well with direct attach storage, virtual servers, mainframes, and thin provisioning.
While everyone else was pursuing SAN based storage virtualization, Hitachi was taking a simpler approach. Since the Hitachi enterprise storage had been doing virtualization services like LUN virtualization and non – disruptive copy, move, replication, and upgrades for some time, the simplest approach to virtualization was to enable other multi-vendor, heterogeneous storage systems to attach to the Hitachi control unit and present their LUNs through the Hitachi control unit to the application servers. This control unit approach to storage virtualization fulfilled the promise to mask complexity without introducing another layer of management. It also enables the aggregation of content and files services, provided by gateways with the block services of the storage control unit to enhance lower level storage systems with the performance and services of the Hitachi Universal Storage Platform.
This product and it’s follow on, the USP V/VM, have been successful with over 9200 units installed since September 2004. Not all of them are making full use of virtualization, a little more than half. But even if they are not being used for virtualization of external storage they are delivering value as a services rich storage system in its own right. With SAN based virtualization, which has no other value other than virtualization, you must make a commitment to make a major modification to a SAN and remap external storage extents in such a way that it is difficult to reverse or replace. The control unit approach can be installed as standalone storage with no modification to the SAN and virtualization can be implemented at your own pace just by attaching external storage to the USP FC ports where software will query and discover the external LUNs, just as they are, with no remapping. If you want to reverse your decision to virtualize, you simply disconnect and connect back to a server where the LUNS can be discovered for native use. As user experience grows, the trend is for more of these systems to turn on their virtualization services.
The question is why aren’t more people implementing control unit based storage virtualization? Especially with the capability to leverage thin provisioning with virtualization to drive up utilization and reduce costs in today’s tough budget environment. I believe the biggest problem is the perception that came out of the SAN based virtualization experience and the fact that the other major vendors continue to beat the drum for SAN based virtualization.
At a recent JP Morgan Chase investor conference, The EMC chairman was asked what he thought of Control unit based storage virtualization. He said that he didn’t get it. Why would you put a storage controller in front of an intelligent storage system like the DMX to dumb it down? Well there could be several reasons. One is to make the DMX smarter by enabling it with thin provisioning, or to migrate its data to a much lower cost storage unit. One of the reasons for doing storage virtualization is to reduce the number of intelligent, expensive, storage units in favor of lower costs storage units without sacrificing intelligence. If EMC could make the DMX smart enough to do external attach, they could leverage their investment in features like SRDF and Time finder, and put more capacity on less expensive external storage.
Its time to take another look at storage virtualization, from the perspective of the Hitachi Universal Storage Platform and its control unit based approach to storage virtualization. You will be pleasantly surprised by what you discover.
The idea for this blog came from Norbert Lanzendoefer, the Hitachi Manager for Presales and Solutions for Central East Europe, Middle East Africa. Thanks Norbert.
