00:08
Hi. My name is Anthony Live Ferrario, engineer on the Cloud Block store team here at pure storage. And today I want to walk you through the architecture behind our cloud block store for AWS product. The first thing that we do is we start in the marketplace where we have an offering that allows you to consume cloud block store directly from within AWS.
00:29
So once you go in the marketplace, you can select that offering and it will provide you with a cloud formation template. Confirmation is going to control the rest of the deployment and lifecycle of your club block store instance, and it's going to be the central place that you'll go to make any changes, upgrades or even to decommission your array, uh,
00:49
later on cloud formation. Then mhm will provision your club blocks, for instance, into your VPC. So let's break down what actually inside a cloud block store. Instance. If you're familiar with pure storage from before, some of this is going to look similar. But some of this is going to look very
01:10
different from the way an on premises pure storage right works. But the thing to keep in mind is The same software runs both on our on premises systems and on our club block store systems. So the first thing that will get provisioned is a set of to control our virtual machines. These are deployed inside of EC two and they represent the front end of your cloud block
01:33
store. So these are easy to VMS. Mhm, mhm. And these are where your management service and your eyes cosy data service are going to be surfaced from and for the convenience of drawing. I've only drawn these lines out of one box. But actually these two boxes work together in an active,
01:58
active manner to provide these services for you. Then what we'll get deployed is the actual storage drives themselves. Now, this is where one major difference shows up between the physical systems and a cloud block store. Instance, these drives are not actually just drives mounted to these EC two instances what they actually are.
02:20
It's their own virtual machines, and we deploy them in sets of seven to represent the similar concept to the way we do shelves in physical systems. Now, as I mentioned, these are also easy to virtual machines. What happens here is these actually all
02:54
connect up both to the primary and the secondary controller. And what these do is they host two different pieces of storage within themselves. One thing is an NV ram volume. This is actually an E b s I 01 volume which, which cooperates across all of these virtual drives as we call them to provide NV ram non
03:25
volatile ram for these, the entire cloud block storage system making sure we can take in and acknowledge incoming I o very, very quickly. The second thing that gets added and I'll demonstrate this one on the far right side here is an instance store volumes mhm or two depending on the configuration of these virtual drive VMS. But this instant store volume is actually where
03:52
the entire working working data set of the cloud block store instance is stored. So as this NV ram distributed, NV ram fills up. It's going to flush out down to these instance store volumes and the same, of course, is done across all of these volumes. That's what happens in terms of general data path between the controllers and the virtual drives. Now, if you know anything about, uh,
04:27
the way Amazon works, the way their services work. You might raise a question here and say, Oh, Anthony, I think you guys are messing something up. There's a problem with your architecture. These instance store volumes. If you lose the virtual drive, you're going to lose this instance Store How How do you guys preserve data?
04:44
I thought you guys were all about six nines. You know, data availability, all that kind of stuff. That's where we have one more key piece of the cloud blocks, for instance, that's different from the way a physical array works. And that is that we actually back end everything in s three. So what you'll see is all the data that gets written down through the controllers and into
05:09
the virtual drives actually ends up making its way out to s three. And that, of course, comes from all of these drives. What this allows us to do is this allows us to handle the fact that we may lose a drive. We may lose two drives, and we're not going to lose any data because when this drive gets re provisioned, s three is just going to feed the data right back into the drive and it can pick
05:35
up right where I left off. Once you have all of this, this is a full layout of a cloud block storage system, and it's ready to provide storage for your applications. And with that, you now have an understanding of our architecture for Cloud Block Store in AWS. Thanks for watching and hope you have a nice day.
