A Conversation About Storage Clustering: Gluster VS Ceph (PART 3)

hey guys welcome to part three of our three-part series on clustering with 245 Drive store inators and open source software clustering software SEF in cluster FS in this video we kind of wrap up the whole story we talk about kind of like hardware best practices configurations when use 3 servers wanted to use just one of something and we introduced the some of the extra bits you need with SEF and our hardware designs that go along with that so without further ado let's get into it let's talk about since we have our nice little stuff monitor sitting here right so this is a product we have 45 drives and it's a sap monitor and it's just a computer that's optimized a standalone computer that can put in your rack it's all solid state it's got what to solid state it's got redundant buck drives lots of memory that's basically what has yet right and it's optimized CPU and memory wise to be able to handle that load and so these are inexpensive put in there and very little power draw and you can put those in your network and then all of a sudden have offloaded your stuff monitoring your MDS you know and and I guess my advantage is got a lower load on the server my data servers can serve data yeah and my monitors kamagra yeah and really the idea there is this like like any good cluster as it grows with you say your new business you need this data you needed high available you know you're gonna scale but you your budgets cut through servers in it because of all these drives well you know what it's fine build your three node cluster and we will call locate everything and then as you scale users as you scale needs you work with our team and we build your cluster with your business so I'm new to clustering and I want to set up a cluster and I'm just crossing that threshold for amount of data and I got Gluster or stuff give me the viewing both of them so yeah what's my minimum what's minimum size how many machines do I do I need three is the magic number so three is the magic number I really should start with three yet yeah and that is from well you need three for quorum and really that's what it comes down to and what I mean by that is high availability if you're gonna have servers all working on the same tasks they all need to be completely in sync with what they're doing if you only have two servers they'll always be fighting with who's actually right if you have a third server watching you've always got corner you've always and that's the idea so set by design literally can't be built any smaller than three notes three monitor services Gloucester will allow less your will allowed to butBut you're not you you were putting yourself at risk for split-brain and some very annoying manual Lorraine is when they they get they lose synchronization yeah and that's what they can't tell which one is correct yes so say the example like I said I've got two servers you're working from the two of them one of them goes down and I don't know for a network clip and then comes back up and now Gloucester all of a sudden sees oh there's that file keep on this server and there's this file on this server but they're different who's right gotcha good and then it does wonderful but if there was a third one there it'd be like I know number one you you fell asleep for a minute I'm Jim oh okay cool replicates itself in its fix yeah so it so suffer Gloucester so what I got again I'm gonna summarize once more what I got is Gloucester simple straightforward file system start off with three servers add them as I want that everything runs smoother and and less problems as it as the cluster gets bigger and bigger mmm-hmm careful of situation without a huge number of I know it's you guys if you want directory listings that may be slow it won't slow down your throughput if you know where your files are your full path your well but careful on your director listens suff if you put stuff together you got this rock solid object store with metadata removed and it's fairly bulletproof and again incredibly simple to to scale Oh yep once you do once you've got the architecture what you want senate-house it yeah exactly it speaking to an IT professional they call it 45 drives and and there that they want to pose this question we've talked about the generalities this so they want to get down rig down to it and they said I'm interested in clustering and I want to use open source so is it south or is it cluster is there always so simple answer to that sometimes it's very simply cluster sometimes is very simply chef and sometimes the answer is that you pick well when do you like better okay so what kind of questions are you asking to try to what are the main features of of the wipey professional situation organization situation that you would okay what kind of file workload are you gonna see out of this thing like are you mooring of a media large streaming type of like you need as much throughput as you can get out of this cluster then Gluster is a good choice for you because you can still build raid underneath everything and you get a huge performance out that way big single file single file that's where that's where you really need it that's where cluster works well and when disaster recovery is paramount above all else if you are very if you're in a situation where you've got this and you need a very safe copy of it elsewhere in the world Gloucester does it better than any other Gloucester raid underneath and other machines in a different location hooked up and using gel replication correct you got it and it's a quick entry so if you're someone who's a little scared of learning something new or simply just don't have the time close your first good choice - okay now what what are the situations worse F is your clear winner so soft that fast is your clear winner um if it's more of a just kind of miscellaneous file store for large companies or not like not large companies but a large file system for a company which is general pdf's here and there say like we said with the metadata thing most normal people don't have files that big maybe a legacy of the business stores all your files in this one folder that's 40 thousand things big maybe you just kind of stuck with some inherent problems with your old file system that you need to deal with that's where stuff is really set fest is really useful for and performance performance is solid but it's not as fast as clusters on the single file transfer on the single file transfer however the the late the i/o the latency the i/o per second out of Asaph cluster is much more pair of parallel than a cluster cluster okay so Brett you're you're a fan of stuff you love so yeah it's it's it's awesome like for all the reasons that we compare cluster fest and South events here both are great I love South fest one of the big reasons why I love stuff is because it can do more than just a file system your same cluster can give you object and block access so it's it's just clean I guess like it just makes it's clean it's easy to use you get one dashboard everything is all wrapped into one and it's I hope everyone soon loves it as much as I do well let's another video sometime about block applications and or object applications sounds good I know I know our sales team would really be excited to tell the world about how we interface with VMware and all that so I feel really well cool thanks Brett appreciate sitting down to chat with you and you know if anybody watching this have any issues clustering here you can sort of put a cluster and come talk to us we would love me love clustering and got a great support team for that'll help you get through this in a very economical way awesome thanks [Music]

Channel: 45Drives

Views: 7,680

Keywords: storage server, storage clustering, storage server clustering, gluster, glusterfs, ceph, cephfs, 45 drives, storinator

Id: uL3zIXQciQo

Length: 8min 22sec (502 seconds)

Published: Tue Oct 09 2018