Realistic Expectations for Thin Provisioning
by Hu Yoshida on Apr 4, 2010
Updated on April 12, 2010.
After posting my blog on increasing storage utilization with Dynamic Provisioning, I visited a very large customer who was in the process of implementing thin provisioning from different vendors. Our Dynamic Provisioning on the USP V was the first to go live. Thin provisioning from other vendors were still undergoing evaluation.
What motivated their interest in thin provisioning was an analysis which showed that their utilization of storage capacity was less than 30%, in some cases as low as 17%. This was a multi-petabyte shop, so it appeared that they had several petabytes of storage that was not being utilized.
Instead of trying to solve their utilization problem by setting quotas and forcing the applications and operations people to manage their storage allocations to the bare minimum, they opted to use thin provisioning technology to increase the efficiency of their provisioning.
I met with the architecture and operations people first and they had a good understanding of what the technology could do and had set realistic expectations for increasing utilization. They realized that not every file systems or data base is thin provision friendly and file systems do not stay thin provisioned if there are a lot of modifications to the file. They did not want to be in a position of ever having to tell their application user that the capacity that they had planned for really wasn’t there when they needed it. They would rather be in the position of helping a user who has just run out capacity in a crunch, provision additional capacity in minutes rather than hours or days. So instead of reserving less capacity than what the application requested, they planned to allocate what was requested even with Dynamic Provisioning. They found that there were other ways to increase utilization without putting their applications at risk.
Since they had high performance demands, they had been using wide striping through concatenation of volumes and software striping. When they would get a request for 400GB they would have to reserve several TB in order to do the striping across multiple spindles. With Dynamic Provisioning they would only need to allocate 400 TB GB from the dynamic pool and the pool would handle the wide striping automatically. Instead of allocating an Oracle data base with enough capacity for three year’s worth of growth, they could use the Oracle Automated Storage Manager to allocate what they needed, when they needed it and provision it dynamically out of the dynamic pool. No need to wait for operations to create RAID groups, carve out LUNs, format, provision, etc. The real value for them was the ability to dynamically provision storage and provision storage on a more granular level. While they did not reserve less capacity for the primary application volume, they still got the benefits of thin copies, thin moves, and thin replication which also saved them time as well as capacity. So they found that they could realistically improve utilization by about 20% with little risk.
Instead of calling this feature thin provisioning, they are calling this over provisioning, because it enables them to provision more storage when they needed it. That to them is the real value since it reduces their risk and their operational costs. The savings in capacity is an extra benefit, but only if it does not compromise their ability to provision capacity when it is needed.
After this meeting I went into another meeting with senior IT management to understand their requirements and expectations. They understood the need to reduce operational costs and the value of storage virtualization to address this cost. They needed help in understanding those costs, quantifying them, mapping technologies like storage virtualization, Dynamic Provisioning, and archive against those costs and communicating the Return on Assets to their financial folks. They also realized the need for services to help relieve their staff of the grunt work that could inhibit their ability to adopt storage virtualization.
When the discussion touched on thin provisioning, one of the executive thought that they could easily increase storage utilization by over 60%. We immediately let them know that their staff’s expectation was more like 20% in terms of utilization.
Like any new technology, there is a lot of hype around the capacity savings that can come from thin provisioning. Unfortunately many industry analysts that talk to senior executives add to the hype. To me the benefit of Dynamic Provisioning is more around the operational benefits of provisioning and the performance benefits of automated wide striping.
Set realistic expectations for thin provisioning in terms of capacity utilization. You don’t need to put your applications at risk of running out of capacity to save a few TB. There may be other ways to increase utilization with out risking your primary data volumes. And don’t undervalue the benefits of dynamic provisioning and automated wide striping.
Comments (3 )
“With Dynamic Provisioning they would only need to allocate 400 TB from the dynamic pool”
It should be 400GB
I have always wondered why HDS does not license “Wide Striping” and “Thin Provisioning” separately. Wide striping has been available on the USP and USPV without a HDP license. The Customer Engineer is required to perform a concat of upto 4 RAID groups though. This is the only downside. We have been using wide striping without a HDP license for our OLTP apps for quite sometime with excellent results
@Nikkel&dime — thanks for letting me know. The post has been updated.