by Liem Nguyen, Director of Corporate Communications — December 21, 2009
Here’s some light storage reading for whatever holiday you happen to be celebrating. Last week I got some questions from Phillip Jaenke, an IT consultant and Unix guy I “met” via Twitter @rootwyrm. He recently worked on a project for a customer of ours and got a little experience managing a Compellent SAN, which prompted him to ask some really detailed questions about our automated tiered storage technology. Since they were a bit too long for Twitter I thought I’d answer him via this blog. I asked Bob Fine, our director of product marketing to help tackle some of them.
Thanks to Phillip for his interest and for sending me these great questions. They are reproduced pretty much verbatim except in a couple places where I edited them for length. The result isn’t quite Click and Clack but you get the idea.
Look for future blogs and video conversations with Bob about storage technology trends. If you have questions or would like to send ideas for future posts, please contact me via email at lnguyen[at]compellent[dot]com or Twitter @LiemNguyen.
Q. With automated tiered storage, what level of control does an administrator have over the ATS behavior? For example, would I be able to control the number of hits on a block required to mark it as a 'hot' block for migration to the fastest tier?
BF: Although we do not provide control over the number of hits per block, we do provide a number of other options for control with Data Progression. These include locking specific volumes to a particular tier and control over how fast we move data between tiers. Our core algorithm moves the most active blocks to the highest tier allowed (also controlled by the admin) which is working well for our customers across multiple vertical markets.
Q. Am I able to control how long these blocks would stay on the fastest tier before migrating downward, and what level of granularity? For example, can I configure ATS so that a block needs to be hit 20 times in a 1 hour period to be flagged for migration to 15k disk, and once migrated, would only stay there for four hours when it drops below that hit rate?
BF: The system is taking into consideration the access patterns for all the blocks in a given volume, and the overall space available on any given tier, to make sophisticated decisions on where to best place the data. This decision is ongoing for the life of the block. You can adjust how long before the blocks move from the fastest tier to a slower tier through our profiles. Right now the options are for days not hours.
LN: The idea of managing the number of hits in a given hour is pretty cool, but the vast majority of customers that talk to us about tiered storage needs want something that’s a lot more “set it and forget it.” The reason they want it automated is they don’t want to mess with the different dials and levers.
BF: That being said, if a lot more customers want that level of control, we’d listen. We consult with our Compellent Customer Council (C3) and Enterprise Advisory Board on issues like next-gen automated tiered storage. What we hear from them is interest in expanding support for different drive types and I/O, not more manual intervention. But customers can definitely work with our Copilot team for some one-off scripting services if they really need even more granularity than we provide today, which is quite a lot already.
Q. Could I also configure ATS so that certain LUNs could skip tiers while migrating? For example, in a configuration with 15k FC, 10k FC, and SATA, could I configure a LUN that would only migrate between SATA and 15k FC, completely skipping the 10k FC tier?
LN: Yes, you can set a policy, or profile, to migrate a volume to a specific tier skipping intermediate tiers. This is customized in the Data Progression user interface.
Q. When a block is 'hot,' how are cache hits taken into account for managing these blocks? Would cache hits count as disk hits from the perspective of ATS, or does ATS only consider disk hits for migration measurements? To get a good bit more technical - is it possible to configure migration to also be based cache hit/miss rate, so that blocks sitting in cache constantly can be migrated to slower disk?
BF: Disk access is the primary factor for data movement, not cache hits/misses. What matters to customers most is disk access speed. Moving inactive data off expensive drives, or moving active data off slower rotating will have the most impact.
Q. Is the Storage Center management interface a separate server, or embedded in the controller? If it's embedded in the controller, is there some way for me to manage an environment with multiple controllers from a single point? (e.g. I have Compellent behind an SVC cluster, and have 3 smaller systems rather than one large one to increase performance further.)
LN: The browser-based management interface is embedded in the controller. All the typical storage admin tasks like mounting a LUN, mapping a server, executing a Replay (snapshot) can all handled via wizards. The design principle we used for the GUI was akin to TurboTax. If you can file your taxes online, you should be able to manage TBs of storage without special training or services.
The GUI makes managing storage for one server at a time pretty easy. For automating storage management tasks for hundreds of servers at time in Windows environments we’ve also integrated Microsoft’s PowerShell 2.0 cmdlets. For non-Windows users we also support a command utility for scripting.
You can manage multiple controllers and sites using a "single pane of glass" via our Enterprise Manager software, which is installed on a separate server. Some of the things you can do with Enterprise Manger are resource reporting across all sites (e.g. amount of SSD and SAS used), administering replication (including bandwidth shaping, live verification), administering chargeback, and creating handy charts that you can hand over to your CFO or CIO.
Q. Does Compellent provide an SMI-S for performance monitoring or array management? (Otherwise known as "can I hook it to IBM TotalStorage Productivity Center, BMC Patrol, Tivoli NetCool etcetera?") In a Tier0 / Telco environment, does it have alarm capability either through SNMP trap or dry contact?
LN: Yes, we support SMI-S 1.2 and SNMP traps. Our PhoneHome proactive monitoring also alerts the administrator of issues and admins can get prompted to do certain tasks like add more physical storage.
BF: We’re doing additional testing with SMI-S 1.4 on enterprise management frameworks from IBM, HP etc.