Response to Correcting PCSE Waste
by Hu Yoshida on Aug 17, 2007
Open Systems Guy, a new storage blog, posed the following question after my post on Correcting PCSE Waste
“I was under the impression that storage virtualization technology spread a workload across multiple storage devices and by doing so, connected those devices to each other. I also thought that it was mainly used to increase the overall used resources of the disks being virtualized… if every LUN in a virtualized environment is striped across several boxes and each box contributes space and spindles toward the workload, how is that less efficient than non-virtualized storage?”
There are many aspects to storage virtualization. One aspect is volume management which is described above. This is the ability to create volumes or LUNs and assign them to an application server. This provides ease of management, having one set of tools to create volumes from heterogeneous storage devices. However, creating volumes or LUNs across multiple storage devices does not connect the LUNs together. Once you create a LUN, whether you create it on one disk, one RAID group, or a stripe of storage across multiple boxes, you have a fixed amount of storage capacity that is fixed to a certain location. You still have the problem of “allocated but unused space”, and the problem of replicating that wasted space across every copy that you make. If you happen to have extra space or lower cost space in another location, you have no way of moving the LUN short of reading it and writing it with software. If you don’t have the ability to partition resources that are shared in a virtualization pool, you will not get users to take the risk of sharing that pool and you will end up with stranded storage. You may also have a problem of connectivity; do you have enough storage ports in the virtualization layer to connect all these LUNs and servers together?
This is the main problem I have with SAN based virtualization. It is primarily a volume manager approach. If we revisit SAN Virtualization 5 years from now we will still see the same inefficient use of storage.
In order to provide a storage virtualization solution that can address the utilization problem, it must be done in a storage control unit platform where you have a large global cache that can present a single LUN image to mulitple storage port processors, and move that LUN image non disruptively across heterogeneous storage systems. The capacity inside the LUN image must also be virtualized so that we can provision only the physical storage that is actually used. Control unit based storage virtualization must have the following capabilities:
Non disruptive data mobility where the physical LUN behind a virtual LUN can be moved or migrated without disruption to the application
Thin Provisioning to eliminate the waste of allocated but unused space
Non disruptive replication of thin provisioned volumes to fixed or thin provisioned storage pools
Virtual storage ports with separate address spaces for scalable connectivity and safe multi-tenancy,
Logical Partitioning for Quality of Service and security
Utilization monitoring tools to monitor use of capacity and bandwidth
Policy based tools that can move data based on policies triggered by time or events or alerts from a utilization monitoring tool
Multiple storage processors that share a global cache, so that the virtualization process does not slow down the application.
The last question that Open Systems Guy posed above about why should striping a LUN across multiple boxes be less efficient than non virtualization, is not a claim that I made. . They are about the same. In other words striping alone does not buy you much in terms of efficiency. It might increase performance by spreading the access across more disks, but it doesn’t increase utilization. I do have a concern about striping a LUN across separate boxes, since each box may have different performance characteristics and, more importantly, different error recovery characteristics. I recommend that virtual LUNs be created in the same storage box and not spread across heterogeneous storage boxes.
Comments (3 )
Thanks for clearing that up- I thought you were referring to inefficiency on the disk side, not the host side. I agree that allocating storage to servers (whether from a storage device or a storage virtualization device) without thin provisioning tends to let those servers squander space that could be usable by other workloads.
A workaround I’ve seen that avoids the storage devoice overhead (from the addressing scheme needed for thin provisioning) is to allocate space to a server virtualization cluster and dynamically allocate storage to the virtual server worloads as they need it. VMWare can do this, and I think some Unix servers can as well. Unfortunately there will always be applications that do not work well (or at all) inside a virtual environment, so in those situations thin provisioning is still needed to avoid wasting space.
“it must be done in a storage control unit platform”
Some comments over on my blog under “Over-allocation vs Virtualization”
I am trying to establish the effect of blogging on company’s communications with their customers/stakeholders.
I.e. does the fact that it facilitates two-way communication
fundamentally alter the dialogue process?
I am doing the research as part of my dissertation for my course MA
Communications Management. It would help me greatly if you were to answer the short questionnaire below and send it to
Let me know if you require any additional information.
Kind regards, Rachel
Student ID 9901912
Questionnaire on corporate blogging – please return to firstname.lastname@example.org Many thanks
Corporate blogging is a relatively new addition to the communications mix. In what ways do you think its evolution has affected corporations relationship with their customers/stakeholders?
In your view what role does corporate blogging play in the communications mix of organisations?
In your view what is the appropriate level of blogging in a company e.g. who (what level) should blog? How frequently?
What arrangements/safeguards (if any) are in place to manage your corporate blog? E.g. are comments censored before they are made live on the site?
Do you have, or have to adhere to any kind of corporate code when blogging?
Could you briefly outline how you feel blogs add or should add to the functionality of a company’s communication with its customers/stakeholders?
In your view as blogging becomes more widespread how can we ensure corporate blogs are effective and useful?
Please comment on the future of blogging and communications and anything else you feel relevant?
Please pass this questionnaire on to anyone else in your company that also blogs on the company blog.