Hyper-converged Infrastructure - Scale Up vs. Scale Out
Transcript
hi it's Keith townser from the C2 advisor.com with another Periscope coffee chat I'm back from delivering the V the Keynotes at both the Aussie and well both the Aussie vmug events had a great time one of the Keynotes was on hyper converged infrastructure as a as a result of that keynote the register picked up uh one of my comments on cost we we won't get into cost but another comment that they picked up on was around scale up versus scale out I was talking about the
advantages of HCI hyper conversed infrastructure versus uh traditional infrastructure so whether it's a conversed infrastructure or bring your own kind kind of hyper Converse or infrastructure environments what's the advantages of ACI over traditional infrastructure also talked about the disadvantages one of the disadvantages I noted was scale out versus scale up hyper conver infrastructures more or less scale out one of the Simplicity spokespersons were there and they kind of or attempted to correct me and I've gotten this feedback quite a bit that uh that when you
say hyper converg tradition uh Solutions traditionally from those companies such as nanic simplivity uh scale don't scale up but they rather scale out that that's fud and that that's thinking Circa 2012 was one of the comments on the register post I'm here to dispel that that it's not fud uh the solutions don't scale out I mean I'm sorry the hyperconverged solutions typically don't scale up they scale out one of the ways that they fix the scale out limitations for hyper conversed is to separate compute and
storage nodes so what does that mean it means that you can buy a base solution let's say a three node solution with both compute and storage you get to the point where you need more storage you can buy a sto a storage Only Solution add that to your or a storage only module add that to your cluster expand the storage nodes out to four clusters versus three and you still only have three compute nodes what does that practically look like that means that the fourth node
doesn't have hypervisor software on there it's just bare metal storage that expands out the storage cluster and you add more storage to your capability so you scaled out the storage array from a storage only perspective and then the other option exists too let's say that you don't need more storage and you you need more compute where you can add another hypervisor node to scale the solution out to add more hypervisor solution so those are still examples of scaling out not necessarily scaling up scaling up means
that give you a example use case let's say that your average VM workload has increased the memory footprint so let's say before when you first rolled out the solution the average solution you 16 that you're average VM used 16 gig of RAM now your average solution has used 30 is up to let's say 32 gig of RAM your average VM so V motioning footprint of a single node so I need to scale up each node in my environment adding a node to adding a compute node
to a cluster doesn't solve that problem it allows me to run more VMS thus scaling out but it doesn't allow me to scale up the average VM workload so I'm sticking by my comment sorry sorry the folks at simplivity uh Your solution is awesome it's a great solution but it's a scale out solution the architecture and I this is one of those things that you know is a pet peeve of mine when it comes to vendors and their marketing teams and their architectures the architectures are
AR architectures I'm not this isn't a disc on ACI on HCI the arch architecture is the architecture if it's scale out it's scale out the some of the limitations of scale out is that if you want to individually scale a a portion of it you're limited they solved that problem in 2012 by giving us the option to do single nod such as single compute or single storage nodes that doesn't make it a scale up solution you're still scaling out that also means that when I want
to quickly deploy a solution if I want to scale up I still got to open up each box pop more memory into it take memory out pop more disc into it it takes time adding a node is much quicker and that's the advantage of scale out so you know the architecture is the architecture I'd love to he hear feedback on the Twitter I'm obviously not watching the Periscope stream as I'm recording but this was a really fun uh uh topic I love whenever I talk about
ACI I know I get a bunch of reac from the vendors a lot of great conversation and a lot of reaction from customers I hope this isn't uh an exception talk to you guys on the Twitter
