00:00
Hello everybody. I'm Jonathan Kearns. I'm senior product measure file services on flash array and I'm Manish Agarwal, senior director for a senior director of product management for fashion defined services. Today, we are going to be talking about flash refined services. We give you our view of how we've built flash refined services,
00:18
how it's different from the competition. So John to get us started, why don't you first introduce what is flashing? Sure flash array is pure storage is data storage platform um that supports both block and file. So it's built out of all. Envy me flash which has data efficiency built into the system through software.
00:48
Things like global storage pool, global de dupe, non disruptive software and hardware have all been built into this platform. Additionally, it takes part into the upper green program here appear which allows you to seamlessly upgrade between different models array without having to migrate data. Um Some of the block protocols they build on here were fiber channel I scuzzy Envy me or
01:10
fabric and DME or TCP and be balls for to list a few and these are all built on top of the data efficiency leaders and the envy me all the all MB MB flash. Do you want to tell us a little bit about how we use that platform with file services? Sure. So before I talk about flash refile services and how that's built differently, let me talk about the legacy systems.
01:31
These are old architectures. Um they were built 2030 years back, there was no flash at that time. So these were primarily built for hard drives and first thing they did was basically retrofit it for flash. And in the process, they introduced a bunch of trade offs and compromises. Then the next thing when these architectures were built,
02:03
they were built either for the block world or for the for the files world. So if they were built for files, you would have an implementation of S M B and N affairs, they would layer on a implementation for blocks. So you would, for example, building fiber channel or ice quasi on top of it. And this introduced an additional set of compromises which basically leads to a house of
02:30
cards. Now let me come over to the pr side. So this is our implementation of file services. This is a new completely new stack that we built from the ground up. It is a first class citizen, the file stack and the block stack basically are built directly on the storage layers. They are not layered upon one over the other.
02:59
And it shares some of the benefits of data efficiency and all envy me flash and it gets all the file stack gets all the properties such as a global storage pool which allows us to build a infinitely large file system. There is no limitation in terms of how large they can be, they can consume the entire array. There's global video, non disruptive hardware and software.
03:20
And obviously they have a green program as well. The way we built it is it's built on a layer of objects built on the objects is a file system. And on the file system runs the file protocols, SMB and NFS. Now by doing this because they are both directly built on the storage here. They're not sharing any abstractions. There's no limitations.
03:44
You're always, when you're managing files, you're managing, you know, natural constructs of files. When you're managing block, you're managing directly blocks, not blocks on file systems. So to take us further a little bit deeper into this John, why don't you walk us through the management model that we built for,
04:03
for files? Yeah. So taking advantage of what Misha said, like being able to have a large file system, not based on your block limitations. We we decided to take a management model that didn't focus on the file system anymore. You don't need to do much there. If it's unlimited size, we can control it other ways. So we decided to go with a managed directory
04:24
based um configuration and management style that we combine that with the ability to use what we call policies and policies are just reusable sets of rules that you can apply to manage directories. So let me give an example of this to kind of help define this. So if I was on file services and I wanted to create my file system and obviously can create more than one,
04:44
but we'll just create one for this. The first thing I will see is I get my first managed directory, we call that manage directory route. So obviously I can apply configurations, this managed directory and I can control the filesystem layer here. But I can obviously wanna, I'll probably want to create more directories below.
05:01
So we'll create a couple directories here. So let's create two directories down. You can create multiple managed directories below a file system. You can create up to seven layers below and they can go different directions horizontally or vertically. Um The idea here is that these manage directories will allow me to do a couple
05:21
different things. I'll be able to monitor them by viewing their space and performance, you know, statistics also be able to apply policies to them. So let's look at what policies can do for us. So we have three types of policies. So we get three boxes, we have export policies, we have snapshot policies and we have quarter policies.
05:51
So with an export policy, what that allows us to do is to say we want to export out via one of the protocols that we support SMB or NFS. So by, by sending up a set of rules here such as like who I want to have access to how I want to handle route or anonymous access. Um Whether the rule is about SMB NFS, I can just simply attach this to one or more directories and allow me to export those all
06:13
out via either protocol or both protocols. If needed. We do support cross cross protocol access. Same thing goes for snapshots with snapshot policies, I can go and configure schedules for snapshots, retention times. And then I can actually take that schedule and apply it to again anywhere I need to. And what this allows me to do is say even if I need to,
06:34
I can take um file system snapshots, but maybe I don't need to, maybe I only need to take something lower down the file system to save myself from consuming snapshot below in my storage. So this will allow you to do that and you, these are many too many configurations. So you can have many export policies to a single um directory or you can have multiple directories connected to a single export policy
06:55
and vice versa for all the rest. Same thing for quotas. If I want to apply that to any level, I want, I can't what this allows me to set hard and soft quotas to any of those directories. So all these are very useful in certain situations. For example, quotas is very useful when you're doing home directory. So if I want, if I have a home directory here
07:13
that I want to apply a sub quota to, I can do that. Maybe if I in, in the world of BMS, maybe I want to take snapshots of just my VMS that are belonging to an NFS data store. I can do that. So this allows you to have greater control over top of your day to be able to efficiently use all the space on the array and not consume more than you need to,
07:33
but also be able to do whatever you want wherever you need to. So that was an overview of flash file services and the granular data management model that we built on top of it that you can use to manage home directories, user shares, shared directories or scale it down to the granularity of individual VMS. And comparing that to a legacy architecture, a house of cards,
07:56
what we've built is a pure unified storage system. Thank you for your attention and reach out to us if you want to know more.
