Take Back your Storage
by Hu Yoshida on Jul 9, 2007
In my last post I talked about where all our storage is going and developed the following chart.
Now let’s look at how we can take it back.
Starting with the data portion, we can reclaim space through deletion or archiving of inactive data elements. A resource utilization tool like Hitachi’s Tuning Manager can identify volumes that have been in active for a time and a policy can be set for deletion of the volume or migration to offline media if it needs to be retained for any reason. Archiving tools for email, data bases, reference data, and Enterprise Content Management are available to archive in-active objects or data elements within a volume and move them to lower cost tiers of storage. The Hitachi Content Archive platform, HCAP, provides archiving of data from multiple archive producers onto a common repository, which enables common search across these producers for ease of access.In addition to providing compliance features like immutability and encryption, it provides deduplication for further reduction of storage capacity. If storage virtualization is available, the entire volume may be migrated to lower cost tiers of storage during periods of low activity. Reducing the productive data portion not only saves space, it can remove contention and improve operational efficiency.
Allocated and unused space is the biggest source of wasted space, especially if it is replicated many times. Thin provisioning is aimed at addressing this element of storage capacity. Up to now, thin provisioning has only been available from some modular storage or filer vendors, now it is available from Hitachi in their Dynamic Provisioning service on the enterprise class USP V. This same feature is available on the HP XP 24000 and the SUN ST9990 V. Currently Dynamic provisioning is only available for internal storage, but will soon be available for external storage that is virtualized within the USP V. Hitachi also makes thin provisioning available on their HNAS high performance NAS system. Tony Asaro has written several blog posts on how thin provisioning works and how it can benefit storage users. If thin provisioning is not available, than a good resource manager that tracks utilization should be used to provide information for allocation planning.
Although it may not be advantages to reduce the number of copies that are made for many business reasons, it does make sense to reduce the amount of storage required for each copy. Thin provisioning is one way to do this. Copies of thin provisioned volumes, contain only the portion of the allocation that is actually used. Other ways to do this is through the use of Copy On Write technology where only the changed data is copied, or de duplication which is a feature that is available from some disk to disk backup vendors like the Hitachi/Diligent Virtual Tape Facility. As mentioned before, Hitachi also provides deduplication for archive copies through its Content Archive Product HCAP. If storage virtualization is available, the cost of copies can be reduced by copying to lower cost tiers of storage. Just because the production data may require enterprise class performance and availability, there is no reason why the copies for backup, data mining, or development/test should be on the same storage, contending for expensive enterprise space, cache, and port performance. With Hitachi Shadow Image and storage virtualization, up to 8 copies can be made, non disruptively, to lower cost external storage located within FC distances from the main production site. Some intermediate copies for distance replication can be eliminated through the use of Hitachi Universal Replicator which uses a revolutionary new pull technology for asynchronous replication and three data center replication in conjunction with synchronous replication.
The best way to recover stranded storage is through storage virtualization and logical partitioning to ensure safe multi tenancy and QoS. With Hitachi’s Storage Services Manager we can identify stranded storage which no one is using. Then with Storage virtualization we can move data onto that stranded storage without disruption to the application. Since the Hitachi architecture for enterprise and midrange can define host storage domains for each user, we can provide safe multi tenancy and assure users that even though they may be sharing the same storage resources, no other user can see or access their storage address space. For users who are concerned about an unauthorized administrator accessing their data or another user impacting their performance by stealing cache pages, Hitachi provides Logical Partitioning to partition off the user’s cache ports, disks, and administration controls. Logical Partioning is under going Common Criteria certification at this time. Without virtualization and Logical Partitioning, you can recover stranded storage for non critical data storage use, if you know where it is, and if you can tolerate the disruption of moving the data. It also helps to reduce cost if you use tiers of storage. Stranded storage on lower cost tiers is not as bad as having it on premium tier 1 storage.
The lead time buffer can be reduced by buying smaller increments of storage, which take less time to acquire and provision. You can buy a 30TB storage unit at today’s price and only use a third of its capacity in the first year, or you can buy a smaller 10 TB system today that looks like 30TB. With thin provisioning, you can allocate the full 30 TB today but only need to buy enough for the first year. Then you can acquire another 10TB unit next year at a lower price and another 10TB the year after at an even lower price. (Remember the Law of .Storage Capacity which says that storage capacity is cheaper next year.) With Storage Virtualization each new storage unit is simply added to the virtual storage pool without disruption to the application.
RAID 1, which requires the 1:1 duplication of storage capacity, has been justified by some vendors for its performance advantages over RAID 5. Again depending on the architecture, RAID 5 can perform as well or better than RAID 1. If the storage cache does not have write protection, than the data in cache needs to be written to disk as soon as possible, to avoid data loss in the event of a cache failure. If the cache is write protected than the writes can be acknowledged immediately to the application but kept in cache until an optimum number of tracks are accumulated for efficient I/O to the disk. During this period, new I/O to these tracks would be cache hits which would be satisfied immediately. So instead of a costly disk I/O for each RAID 1 I/O a RAID 5 I/O with write cache protection, can be satisfied immediately to the application and accumulated for more efficient disk I/O, and provide better performance than RAID 1. Changing from RAID 1 to RAID 5 can reduce RAID overhead from 1:1 to 1:3 or 1:7. Hitachi storage has cache write protection so it can perform as well or better than a RAID 1 system without write cache protection. With Hitachi storage virtualization, data can be moved from RAID 1 on an external device to RAID 5 on another external deice without disruption to the application.
These are the tools and storage architectures that are available today to increase the productive use of storage capacity. If your storage lease is expiring, or your storage has been capitalized and you are looking at massive maintenance charges, you need to evaluate these new technologies before you commit to storage that will lock you in for the next 3 to 5 years.
It’s time to take back you storage!
Comments (2 )
All in all, there is money to be reclaimed as we take back this storage. Your implication is that in the aggregate, there is recoverable capacity in every TB of Spinning disk. This is correct.
There is also quantifiable costs for every TB. Our work over the past 8 years suggests that on average, for every 12 TB of usable spinning disk, there is $1M of net OPEX reduction potential, recovered over 3 years.
Hu’s comment, “Tuning Manager can identify volumes that have been in active for a time and a policy can be set for deletion of the volume or migration to offline media if it needs to be retained for any reason” is greatly over-simplified. I have used Tuning Manager in great detail and cannot imagine the policy one would need to define in order to do this action. I think Tuning Manager is a cumbersome, over-emphasized tool that HDS really needs to revamp.
just my 2 cents…if anyone cares…