Automation on the SSD Tier

Arthur Cole

Adding solid-state disks to your storage infrastructure does wonders to alleviate the bottlenecks impairing application performance. But faster throughput alone will not solve all the problems brought on by the relentless increase in data loads.


The fact is that adding a tier of ultra-fast storage can complicate your efforts to manage data flow, leading to ever-diminishing returns as network managers struggle to ensure data is always available from the appropriate source.


Small wonder, then, that storage-automation companies are racing to max out the SSD capabilities on their latest platform generations. 3PAR, for one, is matching its Adaptive Optimization software found in the InServ F- and T-Class storage servers with STEC's enterprise-class MACH8IOPS SSD, a move that delivers what the companies call "autonomic storage tiering" that dynamically assigns data to the appropriate storage medium at the sub-volume level.


This kind of automation is also finding its way to very finite applications. The recent combination of FalconStor's NSS SAN accelerator and Violin's 1010 Flash memory appliance is designed to add an SSD tier to virtual SAN environments, where it can be used as a high I/O cache for critical application data. The setup provides the benefits of low-latency reads and writes even as larger volumes are heading to longer-term SAN storage.


The need to automate solid-state storage is leading to a reappraisal of SSD performance factors among some storage experts. Adam Day of systems integrator SYSDBA, which represents 3PAR among others, cautions against simple comparisons in terms of capacity or throughput. A more appropriate metric would be how well they lend themselves to automated environments, which should tell you more about whether a particular system will help achieve business objectives.


But don't make the mistake of overdoing an automated infrastructure either, according to Storage Switzerland's George Crump. Many organizations simply don't produce enough I/O-intensive data to justify a fully automated platform and would probably do better with targeted PCIe-based SSDs on select servers. If you do pursue an aggressive automation strategy, make sure your targets are optimized for the type of storage your require -- nearline, backup, archiving, etc. -- and be wary of systems that scatter data across various tiers unless you plan to implement file virtualization as well.


In general, automation goes a long way toward streamlining data center operations and leveraging hardware to its full capacity. But it can also lead to overly complicated data environments if not maintained and recalibrated on a regular basis.


Still, as the range of data platforms continues to multiply due to virtualization, the cloud and other advancement, automation will cease to be a luxury for the well-heeled and will become a necessity for the masses.



Add Comment      Leave a comment on this blog post
Mar 23, 2010 11:47 AM Stephanie Hueter Stephanie Hueter  says:

Arthur  -

Good points on automation (streamlining vs. complication). I wanted to point to Pillar's Application-Aware technology. With an Application-Aware storage system, data tiering is based on the values of various business applications. Tier 1 applications receive tier 1 storage services, while lower tier applications (and their associated lower tier data) receive less intensive storage services. All applications receive the appropriate amount of performance, availability and capacity. By placing only high-priority data on SSDs, it guarantees that only those mission-critical applications will use the SSD capacity. Here is a customer example demonstrating this approach: http://www.pillardata.com/resources/press-releases/2009/10-14-2009-Transportation-Networking-Company-Boosts-Oracle-Performance-20X-with-SSDs-from-Pillar-Data-Systems.shtm Pillar's QoS provides guaranteed resources for certain applications that are high priority for the business.

However, once the automation comes into the picture, it becomes more difficult to ensure the right level of storage resources are going to the applications that need it most. Mike Workman said it best in a blog: http://blog.pillardata.com/pillar_data_blog/2009/10/autotiering-of-data.html Algorithms used to move data up and down the classes of storage in the pool are not very good today. The algorithms use data usage patterns in the past to predict the future. In other words, it takes days to decide to move data. To defeat the algorithm, one just needs to imagine cases where the last X days of use aren't representative of the next Y day(s). It's that simple. High priority data sets are not necessarily consistently the same day in and day out-use patterns are not regular, and even if they are, variations ruin the algorithm's effectiveness.

- Stephanie Hueter, Lois Paul & Partners for Pillar Data Systems

Reply

Post a comment

 

 

 

 


(Maximum characters: 1200). You have 1200 characters left.

 

null
null

 

Subscribe to our Newsletters

Sign up now and get the best business technology insights direct to your inbox.