 What I want to do is I want to rewind back about ten years ago This is this is what I've been a system administrator for 15 years And this is when I was a little system administrator and I worked a I worked at a place in Kingston One of the cool things about being a system administrator is we get can we get a lot of computers, you know broken ones We bring home and we fix them So I had a lot of I had a lot of computers At my house bringing them home girlfriend wasn't too happy about that There's there's one thing that I was trying to do when I got all these computers I was downloading a lot of movies a lot of wares a lot of you know like music and stuff like that and At the time they only had like 200 gig hard drives 100 gig hard drives 200 gig hard drive was pretty big and expensive back then so I got all these computers going in a room and I Wanted to make it so that I could access all the computers As a distributed system, but make it look like it was in one place so I only had to go to one place to see my movies or Listen to my music or whatever So I was like there's not there wasn't really anything out there that could do that on a from a Linux perspective that I could or a Windows perspective that I could Think of right away What what I actually ended up doing was I looked at the way live CDs worked and How they how they merge themselves into memory and make it so that you can actually run a distribution a Linux distribution and Still right to RAM as if it was part of the file system that technology is called union file system or a ufs anyhow, I was I was getting I I found a way to make all these computers work together and Provide me a single point of view for my file system There was a few problems with it. It wasn't reliable if one of the computers went down I would find myself Just missing that data the stuff the system would still work I would just be missing the movies folder or something like that You know or my music or part of my music depending on what part of the the union file system was missing anyways So it wasn't reliable and it was I had to do a lot of organizing so you could only rate to one place one server at a time So it was kind of was it was a lot of micro management and stuff like that so anyhow Girlfriend and I broke up. I gave her the computers came to Montreal got a new job and this is ten years after and The new job. I'm actually working for I was working for a muco at the time helping them build an open-stack cloud Actually before that I was a windows administrator. You guys all love us. I bet But then I got forced on the project The open-stack project and the thing that I was forced to do was learn sef I was like, I don't I didn't really like the fact that I was you know forced into An open-source project that I knew nothing about However, as I started to get you know started to get into it I realized this was the holy grail of storage. This was everything that I wanted ten years ago It's too bad. I didn't bring my computers, but I must say it's pretty awesome this sef stuff It's a distributed storage system. It runs it runs off of something called rados and crush The way I like to view it is rat rat us as like, you know in business. They have policies and procedures policies are there to be defined and then you design procedures to To commit to policies so it's like procedures drive your policy policies like the direction you want to go The way that that's the way I kind of see sef and and it's too too main logical components rados is like the policy and Crush is like the procedure the the way that it goes about defining the the policy So I'm gonna I'm gonna tell you what the the acronym for rados stands for it stands for reliable autonomic Autonomic means self-organizing in case you guys don't know D is for distributed and The OS is for object storage This is a new technology. It's it's it's really awesome. It is reliable. It is self-organizing It's it it allows you to store vast amounts of data solves all kinds of storage issues so anyways when I was When I was at or a mute go and I and I discovered this holy grail of storage I started playing around with it and I was seeing speeds of like that the bottleneck literally if you got a few SSD drives in a Node the bottleneck on this on a node is the is the 10 gigabit network cards If you've got them bonded or link aggregated you can see upwards to 20 gigabits per second So we're talking about very very fast reliable Storage and it's it falls a Linux methodology of do something and do it really well I'm right now. I'm I'm here just to I can't really get too deep into it because I only got 20 minutes But I'm trying to transfer the enthusiasm about this project to all of you guys I don't know like if you guys have computers lying around. It's definitely worth picking up and learning Now I want to get into crush That's how it goes about being reliable self-organizing It stands for crush stands for Better check my notes here That's It's a little tired today crush stands for controlled replication under scalable hashing if you it's Yeah, yeah, so hashing is what helps it be helps it be safe and reliable The the replication also does that the fact that you typically when you start off with a Seth cluster, you'll have you'll have a minimum of three control three Well, no, you're actually what typically you'll have You'll have six nodes Or if you're just doing it if you're just doing an experimentation just to check it out You'll have four because you're gonna have monitors which keep track of the health of your cluster and tell Your clients whatever's trying to connect and pull data from the cluster it tells The clients where the data is which object storage device your data is on So you have a monitor during the sets of three there are not numbers because you need to maintain a quorum Which means that you can't have two or an even number because something Something could happen where An even amount of servers are down There's no witnesses to see who's you know, which which server has control of the cluster So that's why you typically want to have three servers. So if an odd if an odd number goes down You'll still have you'll still have quorum So you could have like if you have three If you have three monitor servers one can go down and you'll maintain quorum if to go down you won't Your entire cluster will be unusable You have storage nodes. They are pretty much just they just host hard drives and They have daemons object storage daemons called OSDs that sit on the hard drive and They are what allows you to store all the information all the vast amounts of information. Oh Sorry I stepped on the cord yeah, so I Also wanted to get into the differences between Seth and Legacy storage like sand and NAS This I'm sure everybody here knows what a sand a storage area network is or a nazi's network attached storage the difference between those two sand is runs at a lower level where you're you'll be able to You'll be able to format a You'll be able to format a hard drive as if as if it's as if it's part of the operating system for a sand for a nazi it's actually Formatted the hard drive for you and presented the storage it through a number of protocols like FTP server message block and so Well, it looks like it's locked So those are the two those are the two Two main forms of legacy storage one of the problems of legacy storage is that you have a you have a single point of Contention so whenever your your client wants to talk to your sand or your NAS it has to go through a controller that Is typically it can be 10 gigabits or a hundred a hundred gigabits 40 gigabits usually 40 or 10 But that's a single point of contention. So all of your communications to your Sand or your NAS has to go through that point and It it's the more storage you add the the more of a point of contention that that becomes With Seth, I like to I like to look at it as terraforming It's a peer-to-peer protocol So when your client actually needs to talk to the cluster it can talk to any of the nodes at the same time So the the the storage capacity and speed scales linearly you add a storage node you get You get 20 gigabits of performance from that from the cluster added Instead of when you add in legacy systems you add more storage it becomes slower the entire The entire infrastructure becomes slower That's so that's one other thing I wanted to go over Yeah, well, that's that's pretty much about it. That's I tried to convey my Enthusiasm to you guys. Do you guys have any questions about stuff?