Can you clarify for me if the GA version has snapshotting -- a variety of the resources talks about it, but I swear I saw a response on the keynote that it would come later.
Sure. We had quite a few questions on this in the live Q&A session but here it is again: Our snapshots are very unique and elegant integrated completely with the rest of the architecture i.e. inline data reduction, in-memory metadata etc. We'll be releasing snapshots in a post-GA release.
As you add new x-bricks you also add new FC ports correct ? So what is the strategy in terms of front-end connectivity, do you zone new ports to hosts ?
Lets say you have one brick zoned to hosts. If you add another brick, you would gain 4 more FC ports and 4 more iSCSI ports. (2 each controller). You would have to add zones per best practices on the existing hosts and/or new hosts now utilizing the box. However, if the same host(s) that was originally zoned to only the first brick, and you added a brick, you could take advantage of using the added brick but the paths for HA etc would not be setup per best practices. The more bricks, better performance for IOPS and obviously greater physical and logical space, with maintaining <1ms latency as you scale. When the system expands, resources remain balanced, and data in the array is distributed across all building blocks to maintain consistent performance and equivalent flash
Arindam or Miroslav can expand on my input if needed:)
Chris Carrieri wrote:
However, if the same host(s) that was originally zoned to only the first brick, and you added a brick, you could take advantage of using the added brick but the paths for HA etc would not be setup per best practices.
can you please expand on the HA part a little bit. Let's say today i have one xbrick and one host is zoned to it. Tomorrow i add another x-brick but my host remains zoned to the first x-brick. Obviously i can't take advantage of the additional paths to the new x-brick but in terms of HA ..what am i missing ?
Once a unit goes from one brick to multiple bricks, IB switches are added to the configuration, therefore the controllers can now all communicate internally. So IOs that come into brick #1 can still spread data across the newly 2nd added brick. However, what I meant by HA, is that if FC ports went bad on the first brick or went down, you are vulnerable to paths until you actually zone the new FC ports to your Host HBAs. So instead of 2 paths per HBAs via each fabrics, you need to make it 4 paths for each HBA. Does that make more sense. Sorry for not being clear.
Hi Chris, dynamox,
To expand a little, I'd say that the general philosophy is to treat the entire cluster as a single storage array. So in general, we want as much connectivity between the client and cluster as possible, and as many paths between the targets and initiators as possible. In the example, we went from an optimal configuration with one X-Brick, to a sub-optimal configuration when another X-Brick was added because the number of paths could now be improved. The cluster will gladly serve the already registered initiators through the new target ports, but in the example given the switches would restrict the paths based on zoning. That would imply that as part of expanding the cluster you should revisit the zoning on the switches and add the new target ports to the appropriate zones.
In general, we want as many paths as possible between initiators and targets. As Chris mentioned, we will service storage requests from any target port regardless of which X-Brick holds the actual data. If the data resides on a different X-Brick from the target port, we'll just traverse the backend IB network for the SSD-side requests. That said, there may be reasons to limit the number of paths available (traffic isolation, security, host OS multipathing limitations, etc.).
Lastly, as already mentioned in several places, we're delivering cluster expansion in a future release targeted for 2014. So while it's great to understand the theory of how things will work, this discussion is theoretical for the near future. We'll work with customers to help make sure they size their initial cluster purchases so that there is little risk of outgrowing their capabilities.
Take care and hope that helps,