Okay, thanks for joining the session today on leveraging NetApp on tap for open source applications in the cloud. I'm Ben Slater, the chief product officer for Insta Pluster by NetApp and uh pleased to be with you today and hopefully we have some great things to talk about. Just starting off uh with the mandatory confidentiality notice and uh general purposes disclaimers and so forth which you can have a read through and then jumping in to the actual presentation. So really what we're here to talk about today is alittle bit of my journey uh coming in to joinNetApp about 18 months ago when this cluster was acquired and sort of um you know coming with the perspective that you know while we all know that NetApp on tap offers many powerful features that it generally is uh positioned in the tra of the traditional world of enterprise IT and perhaps doesn't have a setting in the more newer worlds of cloud andopen source. Um but we've definitely been on a learning journey and uh discovered that there are a lot of benefits in that newer world with ONAP. Um and uh you basically going to walk you through some of that journey today some of the things we've learned. So to turn that into an agenda uh first of all just a bit of more introduction andyou know whoand what is instacluster um and then we got a couple of case studies where we've done a fair bit of detailed work around Postgress on Azour net app files and CFKA on uh FSXN AWS FSXN uh and then a little bit of looking intowhat's next. So firstly about Instacluster. Um so Instaclusterwe are specialists in open source software for high scale and high reliability. Um we were acquired by NetApp as I said about 18 months ago. Um but basically uh what we do is we have a managed platform for ozen source software. Um and you can see themlisted there. U as well as um pure enterprise support and consulting uh around those uh those pieces of software that we work with. Uh we've been around for a bit over 10 years and have you know over 11,000 nodes under management now as well as many more uh covered under support contracts. Um one very unique thing about InstaCluster we not only support the standard cloud environments of AWS, Azour and GCP uh but also an on-remise managed platform uh offering uh which is of course very relevant to a lot of traditional NetApp customers uh and we'll talk a little bit about more about that towards the end. But jumping in now uh to uh Postgress SQL on Azour uh NetApp files for so for those of you that aren't aware Postgress is a verypopular uh open- source relational uh database probably the most popular of those um Instacluster has had a Postgress offering on Azour andour other platforms forquite some time. Um but you know having been acquired by NetApp andtalking to the uh NetApp experts in this we thought there was an opportunity to work with Azour NetApp files and really enhance our Postgress offering. Um so we started that work in January 2023. Um it did require some you know reasonably significant changes to the way we go about provisioning uh new nodes and uh and manage themuh once they're up and running. uh you know things like creating an ANF account then capacity pools and volumes are allthings related to NetApp files um different storage subnets and security audits uh you know a range of work there over several months um and then there was really quite a big uh tuning exercise over many weeks um and you know tuning some Postgress um settings as well as some um some disc settings uh before finally getting to a GA release in June uh 2023. So quitean involved project but with somereally good outcomes uh as we will talk about in a second. So yes, 325% faster and 70% cheaper uh in terms of transactions per second are thereally fantastic headline numbers that we got out of all that work uh that we did there. So uh a very compelling business case and certainly I think uh you know took some of us by surprise oncewe saw those numbers. So just drilling down into that a little bit. Um you can see here uh a couple of graphs showing the um performance results um and comparing you know the blue line that'sthe higher line there um with a couple of other offerings. One is uh the existing insta cluster managed Postgress ona standard discs on Azour. Uh and the other one is another um competing uh managed Postgress that also uses standard discs. So you can see you know prior to this thetwo standard offerings were pretty much identical in performance. Um but once we introduced ANF to the picture with exactly the same uh you know instances in terms of processes and memory and those kind of things um the throughput in terms of transactions per second that we're able to achieve you know up went up by as it says there 167 or 325% depending on thescenario you're talking about. uh you know um so obviously the it's there's been more of an impact in terms of the uh the read performance than theright performance uh you know based on those things but still uh you know a remarkable impact uh either way and then once you work through that uh the cost of the different uh types of storage offerings you get to those um price performance savings that we mentioned on the previous slide uh we're not done there though with Postgress on ANF um there is uh still more uh we can get not so necessarily in terms of performance but in terms of functional benefits out of ANF. So one of the great things that ANF does like on tap is uh fast snapshots uh you know no overhead snapshots um and so we've actually implemented that already and are now going to be using that to do database forking so ability to just make a copy of your database andspin it up uh you know withinminutes. uh cross regiondisaster recovery uh and then also adding more disc scaling options. So uh ANF also has um great scaling benefits in terms of being able to quickly scale up and down compared to the other uh storage offerings on Azour. Uh and we'll be expressing those through our offering as well. So the next thing I wanted to go on to was uh Apache CFKA on Amazon FSX for NetApp on tap which is a bit of a mouthful. Um but so CFKA for those of you who don't know is a distributed horizontally scalable streaming data platform extremely popular uh for all sorts of use cases of you know effectively a bit like uh message queuing message in ceue them up messages out but uh you know a streaming fashion uh our current offering runs on EBS andlocal discs so uh you know ephemeral discs in AWS and again wethought you know uh there's a some good possibility of getting some benefits out of uh FSX here. Um CFKA is a very different beast to uh Postgress. It's a distributed application uh you know which Postgress isn't. It's streaming rather than um rather than you know ad hoc access. Uh tends to be more appending to files rather than random reads and writes. Um and in addition to that, you know, FSXN has very different performance characteristics to Zuluet app files. So we knew we were going to learn a lot uh going into this as a very different uh exercise to what we did with Postgress on ANF. Uh and we did learn a lot and we'll get to the outcomes. So what we were what we're targeting was the reasonably common use case in CFKA where you have uh themost recently written data is fairly hot. Uh it gets read pretty much straight away by a consumer. But you might choose to store quite a backlog of data, you know, days or weeks worth of data, evenlonger. uh just in case you need to replay and refill a system uh where you've had a corruption uh or other need you know f further down the line um and so we thought that the um tiered storage and data compression capabilities ofFSX for ONAP uh would uh be valuable there um and you know did a lot of work in those areas uh we also think that thesnapshot copies and forking which we talked about with Postgress which are also available on FSXN uh willbe valuable in CFKA as well and something that'sjust not typically available in the CFKA world. So put quite a lot of effort into then this as well um similarlevel of effort into you making our uh provisioning system able to work with FSXN and then a lot of tuning. Uh we did see a pretty dramatic difference between where we started off with the tuning andwhere we ended up. uh most of that benefit was from following theNetApp NFS best practice advice and applying that in the uh FSXN uh context. Uh there was also some CFKA tuning uh along the way as well, but somepretty big changes there. Um just to illustrate where we were aiming uh forthat u and what I said about is you know on the graph on the bottom right there you can see that um effectively as storage gets bigger the uh cost of fsxn grows more slowly thanthe cost of EBS due to the compression the possibility of tiered storage and all those kind of things. Um the other factor though is that the IOPS and throughput are generally more expensive onuh FSXN than they are on uh on EBS. So we have to get thatbalance where thecut over point where FSXN becomes uh becomes cheaper uh thanEBS theother option that's available.Where we ended up at this stage is we didn'tquite get there. We've uh generally discovered that although we can get thatuh reduce [clears throat] cost as we get bigger, we haven't been able to get uh to the same performance levels yet. Uh we've made some big strides and they're sort of somewhat similar. Um but we're going to sort of continueworking on this one which I think you know just illustrates and the reason we still included it in this talk is because it does illustrate um you know when we start working with this stuff the need to understand the differences between an ANF and an FSXN although they're both using ONAP they are quite different uh environments and you know and targeting different use cases. Um so uh that's sort of where we are today. uh whatare we working on next? Um so first of all in the cloud area uh we are going to keep working in the Apache cafka space and look at uh we you know those same performance benefits that we got with Postgress uh maybe we won't get to the same quite level but I think we will be able to again see price performance for CFKA are targeting a different use case targeting the really hot CFKA data um because ANF doesn't have tiered storage uh we still you know haven't given up on FSXN uh the snapshot and forking benefit is there uh we'll continue working uh with the FS FXN team onthe performance tuning and see where we can get to there. Uh Apache Cassandra is another big application for Instacluster and we'll be looking on Cassandra on ANF for our Azour Cassandra customers again with a very hot data use case. Um and then uh Google NetApp Google Cloud NetApp volumes obviously a very new release from uh Google and NetApp. Um we you know from what we've looked at so far we think that'll look probably quite similar to um to ANF. Uh and we'll definitely be you know moving there with Postgress and CFKA and so forth inthe near future. Uh the cloud isn't the only place that we're working with NetApp storage. Uh we have also as I mentioned up front u you know as part of our on-prem manage platform we've recently implemented uh storage grid uh as the standard backup location for that. So that uh that's you know sort of out of the box storage goods devices are supported as your backup location. Savework and time there. Uh we'll be moving to standardize ontap as the standard search storage provider. So having testing and certifying of specific versions and devices withthat NetApp storage um and we're also working at the moment on a Kubernetes-based offering for uh on premises environment and we we'll certainly be looking at NetApp Astra uh as part of that as well. alo a very integrated story with the netapp offerings. So just to summarize that up into some key takeaways um you know there definitely are significant benefits uh from you know NetApp on the world of open source uh and distributed applications uh and you know in the cloud thecloud provider the hyperscaler offerings of um uh are a great way to access those benefits. uh you do however need to know your application and your use case and the characteristics of the particular ontap variant in the cloud to understand how those benefits apply to you. As you illust as we talked about we went down a bit of what turned out to be nota uh a um fruitful path in one place but in the other place wehad some really uh you know more uh amazing benefits than we thought we would get when we started off. Um and I guess you know the final one is there is a lot of hard work in some of this stuff. uh you know what we do at InstaCluster is do that hard work for our customers um so that you know we do it once and multiple customers can benefit uh and you know we we're taking a look care a lot of that work for you if you want to get those benefits with the applications that we support um just a few resources uh that you can access there's a white paper on postcripts and ANF the NFS best back practice guide that I mentioned uh and instaccluster.com is the inster website which has you know drills down more into all of these uh technologies and those kind of things, we have a very active and healthy technical blog uh and you can uh access a free trial there of the incluster services. And just finally to finish off, stay connected. Uh there's all thecontact details there including my email. Uh if you do have any questions or want to get any further information, please uh feel free to reach out through one of those channels. And thanks for your time today.