Share this post

August 30, 2015

Atlantis USX SmartSnap – Snapshots for the agile IT infrastructure

Priyadarshi Prasad - Atlantis


Snapshots are one of the most widely used data services within storage. They have been around for a long time and most storage vendors support them. The traditional vendors however had designed their snapshot capabilities when the IT environment looked very different. Storage was primarily disk-based (HDD) and technologies like deduplication were still immature for primary storage environments with a big tag “try at your own risk” mentioned in user guides (usually the guides would first mention the performance penalty and then render the feature all but useless with best practice).

In the software-defined storage world, we have a new opportunity to re-invent how technology is used and deliver exactly what IT needs, leveraging our (Atlantis’) enterprise proven de-duplicating filesystem– without the limitations of legacy code and monolithic storage frames. So when it came to designing Snapshots for Atlantis USX (and HyperScale), we asked ourselves what would be the most valuable solution for our customers? And so we set out to provide the best snapshot implementation in our industry today. We focused on four key parameters:

Fast and Performance optimized

Traditional implementations of snapshots are essentially copy-on-write (CoW) and they worked ok when performance was not a primary requirement and workloads were predictable. However in the vastly growing application estate there isn’t an IT shop that can relax with predictable workloads. That’s why it was obvious for Atlantis USX SmartSnap to implement redirect-on-write (RoW) snapshot, suitable for the most performance intensive applications. In this age of solid state storage, an added advantage with RoW is that it reduces writes by 50% keeping the now ever-ubiquitous solid-state media happy with less wear.

Traditional Copy-on-Write Snapshot

  1. At the time when a snapshot gets taken, the metadata is copied to the snapshot location but still points to the original data location.
  2. When a write update comes to a data block:
    • The original data block is read
    • The original data block is written to a separate location.
    • Snapshot metadata now points to the new location (preserved state of data).
    • The original data block is then modified with the new write update.

So the CoW update requires one read and two write operations.

Atlantis Redirect-on-Write Snapshot

  1. At the time when a snapshot gets taken, the metadata is copied to the snapshot location but still points to the original data location.
  2. When a write update comes:
    • The original data is read
    • The original data is modified with the write update and written to a separate location.
    • The active file system metadata is updated to point to the new block location while the snapshot metadata keeps pointing to the original location (preserved state of data).

So the write updates require only one read and one write operation, i.e. 50% less write overhead.

Efficient with Deduplication awareness

With Atlantis USX, de-duplication is always inline, can’t be turned off, and is done once-and-forever. This aspect of our de-duplication is incredibly helpful for snapshots. In the comparison above, I have implied that a host write results in a single write to the file system. However, if the incoming host write can be de-duplicated, then that write results in zero (zilch/nada) writes to the file system. It is simply a metadata update and very efficient – sequential and super-fast. It is important to note that the de-duplication awareness of SmartSnap I describe above can only happen with inline de-duplicating file systems. File systems that do post-process dedup will, by definition, have to write incoming write requests incurring the performance penalty this overhead continues when they snapshot the filesystem.

Flexible – With both Volume and VM level snapshots

There will always be a debate on which snapshot technique is better – Volume or VM level. Each has their own benefits over the other. Volume level snapshots are management efficient – it is far easier to manage a few Volumes than it would be to manage potentially hundreds of VMs individually. However, Volume snapshots are coarse and that’s where VM level snapshots can offer more granularity.

This is why Atlantis USX SmartSnap offers both capabilities – Volume level and VM level snapshots. Our VM level snapshots is a direct result of our investment in VMware VVols support – Atlantis is the only software-defined-storage or hyper-converged appliance to be VVol certified. Once you enable VVol (a single-click, no-install operation), any VM snapshot you take in vCenter is offloaded to Atlantis USX by VMware. USX then uses a feature called Atlantis FastClone to create an instant zero capacity snapshot of the VM.

Easy to Use

Snapshots can be enabled by default for all volumes right at the time of their creation, or on a per volume basis as it is deployed. You can also enable snapshots by default and then disable it for certain volumes. SmartSnap provides a fine grained scheduling capability allowing a SmartSnap to be taken on an hourly, daily, weekly or monthly basis. A Volume can be rolled back to any snapshot in the tree – all de-duplication aware, space efficient, and simple.

So how do I see the competitive landscape? My best understanding of the various snapshot technologies is depicted below. As always, if I got something wrong on the competitive side, please let me know at @AtlantisSDS, or @Priyadarshi_Pd.

Current rating: 4.3 (6 ratings)