Storagebod Rotating Header Image

Storage Vending Machine?

Self-service storage provision every now and then comes on the menu at work and I'm wondering if anyone has ever seen this work on a large scale? Has anyone tried it in a complex multi-tiered storage environment.

When I say self-service, I mean the user is presented with a series of questions and at the end of the process,the storage is automagically provisioned and made available. This sounds all very well and dandy but what really concerns me is how this would be done practically and I'm wondering if anyone has used tools or even built their own? I think things like wide-striping make it more feasible but I think the questions to guide the user may elicit blank responses.

For example, how do you ask a user to define performance? Ask most users to define their workload in IOPs and they will have no clue. Ask them whether the workload is going to be random I/O? Another blank look I suspect. Ask them whether the workload is high, medium or low; you'll get an answer, wouldn't bet your life on it.

So let's think of the questions which really need to be asked and the sort of detail we need to make appropriate decisions? 

  •  Space Requirement – Today and 18 months hence?
  •  I/O density – (iops per Gig) and does it increase with the amount of space?
  •  Availability Requirement?
  •  Sequential/Random mix?
  •  Read/Write mix?
  •  Peak load time?
  •  RPO?
  •  RTO?

I am quite happy with the concept of automated provisioning tools which can be used by storage administrators and I'm certainly happy with the concept that my teams can manage more storage per head. Automation is a good thing but are we really ready for self-service provision of storage? 

Perhaps if we had a one size fits all uber-tier? Aha!! XIV here we come!!! Or perhaps, self-aware-storage!


3 Comments

  1. marc farley says:

    Another excellent post, Martin. I don’t think end user self-service storage in large enterprises makes sense for the reasons you give. I’ve seen it work in small businesses where the environment was mostly defined by file sharing, email and one or two line-of-business applications. Data protection in these accounts is typically automated backup where the self-service operator is mostly responsible for changing tapes on schedule. “More advanced” features such as setting up and restoring from snapshots tend to be a bit difficult for people who don’t work with storage regularly.
    Otherwise, automated tools designed to make storage professionals more efficient are far more valuable for enterprise environments. Anarchy is not a great model for storage management.

  2. I looked at delivering this in one customer for test/development. Unfortunately monolithic arrays make this really tough due to their inflexible configuration model. Some modular arrays are better, but there’s always the eternal question of making production-type changes automatically. Most financial sites are locked down so tight these days you can’t f**t without change control. Oh, and if you do dynamic provisioning, you need to do dynamic decommissioning. Imagine the problem if you had one small bug in your decommissioning code….

  3. Ian says:

    Yup tried this 4yrs ago – and even had a go at an XML’like’ schema for it’s definition… Gets v complex v quickly… Also need to consider things in the usage / service chain (eg mutual inclusion/exclusion’, accss points & methods etc.).
    We must be able to get to this point (and maybe object stores – eg Caringo/Atmos etc are a way there)

Leave a Reply

Your email address will not be published. Required fields are marked *