Interop Session – Object-based storage – the next step

Posted by gminks in interop | Tagged , | Leave a comment

Here are my (mostly) unedited notes from the panel about Object-based storage.

Vanessa Alverez also hosted the  panel. The panelists included representatives from Cleversafe, Amplidata, Caringo, and Rackspace.

  • Question: Intros, and describe object-based storage
    • Amplidata: they do OOS – Optimized object storage
    • Caringo:  store lots of unstructured data (file sys can’t handle the amount)
    • Rackspace: Cost to store data is going down, cost of keeping data is low enough to keep massive amts of data around just in case. Next decade should have 50x amt of data. Can’t go on block system, can’t scale out NAS to accomodate. Most data doesn’t need high end features of arrays, not most cost-effective way to store
  • Question:  Adoption is still nascent, so what is your customer’s value prop to move to object storage
    • File systems have inodes, blocks, rigid, not geo dispurse. Unstructured content makes this hard
  • Question: What is the role cloud plays into object storage?
    • Nature of the content. Digital music, video, cat scan, surveillance. Just blobs of data, not a fit for file system.
    • Driven by application
    • Cloud is accelerating the adoption.
    • Object is driven by data type, and by all the hand held data-producing devices available
  • What is the Rackspace value prop?
    • More use out of SAN/NAS platforms, you may be keeping data that doesn’t belong there, also DR/BC.
    • Cloud-based object storage. ISV software companies push to cloud storage as offsite data storage
  • Question: What are challenges storage admins and IT organizations, and the nature of object storage?
    • It’s about the data. interfaces are a challenge and technologies are evolving.
    • A single tv show generates a pedabyte of data. They (entertainment ind) doesn’t understand object based – education req’d
    • Legacy application issues too (client-server apps need to be updated)
    • Hybrid system – db/filesystem. For exp, x-rays. If you have to get it, better to have object id for DB for actual images. Makes the system you have more efficient.
    • Object stores don’t want to turn into a relational database
    • History & inertia (challenges). We all organize and think in files and folders. Apps speak CIFS and NFS, which are file-based
  • Audience question: ability to (keep with) racking servers is the limiting factor to deploying
    • Rackspace – based on OpenStack. Can role in cab full of servers, use Dell OpenStack makes it super easy to set up object storage
    • No head unit scale-out limitation
    • No worries about partitioning, keep adding and it will re-balance as you add
    • Scale out NAS systems. meta data mgt problem, multi headed concurrency problem.
    • Scale out systems run out of inodes. These systems weren’t designed for massive amountsof data we see today
      • Scale out was designed when 100TB was gigantic. Not designed for massive data
    • Audience question: How do you back up object store?
      • Software copies data wherever you say to replicate
      • Versioning on object history. As you look at object store and that is use case, find software that does it
      • Enterprise backup vendors are starting to look at HTTP rest backup solution
    • Question: Pipe size recommendations for network so things can be replicated?
      • Driven by use case. Object design is a lot like network design, not a lot of limitations of how big it can be -how much data, peak loads, etc
    • Question: Storing is one thing, but what about moving the data around
      • Object limitations..
      • Object vendors are looking at use cases w best advantages.
      • Work backwards from requirements
      • Competition is scale out, tape, or cloud. app and use case driven
      • hbase runs on top of new file system. People using this as object store. Some people are using Cassandra for this. Differences between what object stores and distributed key value stores (Cassandra, NoSQL etc)
    • Question: SOAP, REST, are new protocols to storage industry. What skill sets need to be learned/relearned in storage?
      • Learning curve – change mindset from files and folders to tagging object
      • Getting car from valet rather than remembering what floor and spot you parked
      • More programmatic, commands that are “put” “get” “delete” style commands. Small list of commands available. Simple interfaces
      • CDMI – SNIA standards. De facto standards – eg OpenStack, S3
      • Dev community is building utilities to work with systems.
    • Audience Question: Is cache a benefit to object-based?
      • Anytime you bring cache we’re willing to do biz. (hahah storage jokes!)
      • Object – large streaming data, benefits to cache lessons
    • Audience Question: Security models
      • These are different w object based storage. You will have keys to object, users keys, etc. Not standardized across object stores
      • Tagging objects and forcing security – industry is working on it, not standard
    • Question: hybrid environments – how to transition?
      • Use case 1st, what apps to use (Use case seems to be a trend…)? Sharepoint supports blob…can you drop it into a store. If not, you will have lots of development work
    • Audience question: What are the implications for digital preservation?
      • With object-based storage hardware flows under data (which is guaranteed) and can change but data won’t
      • Object makes it easier to do things to make data last forever.
      • File-systems have helped it, object systems have check sums over time that can be verified
      • I heard a new term in this session – bitrotting
    • Question: what is the 1st step to make the transition?
      • Lots of people use public cloud to start. Cloud storage gets expensive, or the corp feels that data shouldn’t be public. So they suck the data into a private cloud, use public for DR
      • 2 usage models – natural fit for those who use public cloud. Sometimes it embedded into archiving solutions, etc
      • Smaller enterprise can have more real-time set up for DR (as opposed to tape restore). cost is comparable, easier to recover
      • Pick storage heavy object app, change it behind the scenes and see if maintenance /management windows are shorter, or if  users comment on better performance
      • Roll out on existing app to see if it is better than what you were using
      • Admin & object store owners impacted the most – even more than users 9BTW drop box is S3 admins store)
    • Question: Does network team and virtualization team need to be involved?
      • A object storage implementation will touch as much as the ecosystem as application does..
      • Object store shouldn’t give you more
    • Audience question: What about vendor lock in, is open source mitigating vendor lock in?
      • Lock in depends on your use case –

Leave a Reply

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