Modern Day Primary Data

Auto-generated captions from YouTube, not hand-corrected, so names and technical terms may be imperfect. The video is authoritative.

all right you've made it to the last session of the ctl advisor virtual seminar we have karen lopez moderating a panel of pretty uh a pretty interesting set of characters from annie banner who has had over 30 years of experience writing software defined storage bits he is the godfather of iscsi in vmware vsphere we have ray lukichi who's the founder and ceo of silverton country consulting who's also my fellow co-host on his podcast great beerus on storage he's been a in a street analyst way longer

than i can account for and then there's enrico senoretti who is the lead analyst for infrastructure and storage over at gigahome they're about to get into it this is uh at parts hilarious but always entertaining and insightful commentary around this aggressive blog post by min io which called the object storage the new primary data i took a fist to that uh claim because i know a lot of people who have primary storage outside of object storage and continuing to grow and continue to make more of

it so i kind of lob this into the team to talk about primary storage what is it what isn't it and how do you make purchasing decisions based on those definitions completely entertaining session hope you enjoy it thank you so much and i welcome all of you for joining us today we've got this abstract that talks about primary data and primary storage make some references to a video interview that keith participated in and some announcements so we're going to talk about some of the futures of

storage and high performance computing as well as i'm going to be asking our panel to weigh in on what they think is the future of storage and high performance computing but let's get started first and i'm going to start with enrico here enrico what's your definition of primary data and or primary storage so uh primary storage is uh where your primary data sits i mean okay that was a little bit of a cheat so but i'll let you continue no well i mean so in the

past we were thinking about the structural data as your primary data okay so databases uh your sap data your you know your erp in general or whatever your article but it's true that today we are producing different types of data there are even a lot of machines producing data that are very valuable for our organization for you know they are really mission critical sometimes and and they are not stored in database or not in traditional databases and uh so the type of storage that you use

it could be fi could be object it could be a block it's your primary storage depending on the data on the data that you store in it so maybe we can talk about the characteristics of primary storage in terms of reliability availability etc etc but actually the main definition i think today is you know primary means critical and that's it ray what's your definition i agree with enrico to a large extent i mean you know primary data is the data that your critical applications for your

enterprise or organization use and and primary storage is where you access that data from it can be block file or object and any of those are are valid uh approaches to those sorts of things i think the challenge we have today is that you know with with buffers and caching and you know direct access storage as well as networked access storage the challenge about where that data resides is is much more difficult i think it can reside pretty much anywhere but you know when it actually

is processed it's processed in memory of the application so how it goes from where it was to memory is uh part of the question that needs to be answered here i think i think i'll leave that at that okay and andy okay and i'm actually gonna slightly disagree with uh with both of those where i i pretty much agree with the idea that primary data is the the primary set of data that you work on databases objects uh i i always like to include virtual environments

any anything that is is the data critical to actually running your business where i'm going to disagree is that primary storage is uh actually is some the place that you store data and the the difference here is that it's primary storage is not a chunk of software primary storage you actually need some way to actually hold that data whether it you know cloud storage is fine or physical storage inside your own environment but i don't i don't like i don't agree with the idea that primary

storage can be served up by a lump of software are you crazy are you nuts the whole world has moved to software to find storage how can you say it can't be served up by software you are crazy no i i i fully agree that there's software there's always been software in front of storage there's always been software in front of storage that doesn't define storage yeah i mean i've been doing i've been software defining storage for 30 years me too but at the end of

the day you can't you set the h inside the h word here you you can't actually store bits in software okay media you need okay you need the full infrastructure you mean i mean you know go back to where some of this discussion started with i don't consider middleware to be primary storage so we at storage field day in a tech field day we talked to a handful of different middleware vendors and yes there they offer valuable services for your primary data but they are not

primary storage they they are middleware that sits on top of your primary storage interesting i'm just going to slide in with a side story i once worked for a national retailer in the us that indeed kept a ton of data in their software like right in the code the prices for all their items all their sku is all the distance there you go it's all software because the two guys who originally wrote their point of sale system had no software background but they knew a lot

about video store rentals and so that's how they they didn't need a stinking database they had there you go but at the end at the end of the day their program existed somewhere it didn't yeah it didn't just sit there and ram the entire life like andy andy talk to me about what what you consider middleware in this space oh yeah good i mean it's well since keith actually brought this up as a discussion about me and i o claiming that their primary stories ah let's

not use vendors names okay so since okay but so that's what i consider middleware middleware is a a software presentation layer that gives you your your your access to your business device in the world every software-defined storage system in the world every system out there is essentially middleware between you and the disk or you and the ssd or you and whatever your platter out there and memory every one of them but in lots of cases the the middleware actually comes with the primary storage necessary to

serve it are we oh so if it's separate it's different give me a break no it is different yes it is different no it's not oh wow i love this um so one of the things i wanted to throw away i'm sorry but primary primary storage has to be able to store bits okay but that doesn't mean it can't be middleware storing bits or appliance storing bits or software sitting on a server someplace storing bits and direct access storage all those things store bits andy no

they don't they they handle bits they pass bits from one place to another they do not actually store the bits i think we're getting into a philosophical discussion because i've sat through plenty of what is a customer discussions in my life and that's where we're going down there's no end to that but i appreciate so much the different points of view one of the things while i was researching this topic i came up with because it was triggered a little bit by a discussion with academics

at a research university is that researchers use the term primary data to mean the data the source data that was collected and has not been altered or enhanced in any way and secondary data is data that is the result of analyzing it doing compute on it enhancing it so just throwing those terms out there so now i want to say that that's really different that means like like don't let me shock you guys that researchers use completely different terms when they talk about compute than we

do um but the only reason i bring this up is of course if there's primary storage there must be secondary storage and tertiary storage like i kind of understand the difference between primary storage and secondary storage but i'd like to hear you guys tell me why do we need to make that distinction of primary storage versus secondary storage well if i can start here i mean so your primary storage is probably the storage you access the most anyway so you access it more frequently so there

is also a performance component to this to your primary storage anyway even if we are talking about object storage that is notoriously slower than block and maybe file so there's a way of archiving data cold storage well i think the distinction is much more complex to do today because you have tearing you have you know everything is hidden behind the scene so even if we are not talking about uh um you know sophisticated mechanisms to archive data and maybe uh you know having a secondary system

to move data away and make it searchable or you know even long-term data like tapes or whatever we're talking about much more complex infrastructures today than in the past and and to the dealing the border that uh that there is between primary secondary and thirty storage is very complex to define i think about there are some vendors today primary storage vendors and again and yes it's an appliance and where you store your data but in the beginning they can do operations to move uh you know

snapshots and and you know blocks that are less used to object storage one can't work without the other and yet so they they in uh from from your point of view you see the front end and maybe the frontend is the latest nvme technology or whatever but you can't separate the block store or the file store from the object storage in the backend because some of the information we are moved to to a secondary tier okay and you do this mostly to save money to have

a you know an instant backup or many other reasons but actually so primary secondary and tertiary storage is a little bit more complex to define it in the past depending again on the use case and how odd or critical is your data yeah yeah i completely agree yeah i saw you nodding your head um because so my follow-up question which i think you're going to go to ray is that do we make these distinctions for cost for environmental impact why do we need to make these

why isn't everything just historically historically primary secondary and tertiary were you know effectively archive layers right so tertiary would be very very cold storage you very rarely referenced at all and you never updated secondary was a little bit more updatable a little bit more accessible but again not very quick to access and and and lower cost each of those tiers is intended to be lower cost but slower performing and lower you know all those things so yeah it's it's a more greener environment so it's more

sustainable kind of view of the world if you can move most of your data to secondary or tertiary storage you're talking cheaper and and more economical and more or lower energy consumptive storage right and i you know i do agree with those definitions with i mean what i was going to say is you're kidding me it's the first one you've actually agreed to primary storage is the storage that you need to be able to access immediately at a moment's notice um you know it's it's what

you're what you work on every day it's what your customers need every day it's the type of stuff that that needs to be at your fingertips and the you know secondary storage is typically something that can be accessed fairly readily but is not going to be there as quickly is not going to be able to deliver it to you as quickly and it's um you know the the secondary tertiary levels are the places that you need to be able to refer to uh you know if

if for some reason data is lost or you need to be able to show some accountability for how you got to your primary data in the first place that's interesting and it's i mean it's it kind of turns your researcher analogy on its head because the exactly yeah i mean the idea that your primary storage is going to be what's what's at hand right now and your secondary and tertiary storage are going to be able are going to be the archival references that you can potentially

say this is how we know our primary storage is good yeah i i i remember from just to name drop here my visit to cern um that they at least the time i was there which was probably about seven years ago they relied heavily on tape backups as secondary sherry storage not just for cost savings but because in they believed that it was better for the climate to make rarely access data not to be on a drive on a machine that needs to be turned out

now they had to provide environmental protection for the tapes but there besides the cost savings of not having it online was the savings of energy usage which i think we don't talk about that much well actually in many enterprises it has become you know one of the major reasons to change you know storage infrastructures i mean yeah hard drives are really really power angry compared to flash drives now and with flash drives you can use techniques uh optimization techniques that you would with uh you know

address drives are very difficult to implement so uh it's true that there are many vendors now focusing on qlc so the latest [Music] storage storage intermediate technology just to you know lower down power consumption make it more efficient smaller also to have a less uh you know footprint in the data but just like just like tape you can you can power off disk drives and you can power off uh ssds and that that technology has been around for a while the the difference with uh with

ssds is that you don't actually have the spin up time to actually make them accessible that you know you can apply the power to them and have them accessible almost immediately right right right the homemade thing i have enjoyed reading vendors sustainability reports that seem to show more than just hey we went out and had this report commissioned to show that we're green but where they're they're actually talking about their design and architectures that are you know sending some thoughts into making things more sustainable for

their hardware even you know the the problem so if you look at a large organization and they are still thinking about uh you know silhouette uh department or sealer teams working on different applications each one with a single budget then it's really difficult to talk about uh about you know environment and sustainability etc because because they are wasting a lot of resources anyway so even if they are purchasing the best array in terms of power efficiency they are using just the fraction of it okay so

the mentality should change uh you know at the top and say okay you are going to share more resources you are going to uh use the resources better and then you can start thinking about because even the uh the discussion about tapes yes tapes in terms of power consumption when they are you know stored properly power consumption is zero okay but you can't build as tape infrastructure if you have just you know hundreds of terabytes of data i mean doesn't make any sense you need beta

bytes and petabytes just to stop thinking about it yeah yeah and that also impacts data quality issues as well like i know on my side of the world grant driven research or silo driven projects leads to a lot of data waste and duplication of data like this whole concept of centralized versus distributed governance and control impacts all those things and one of the other things that came up in the interview that keith did was sort of a mention a little bit on the side that we

have different storage requirements for what i'll call transactional or operational workloads versus analytical workloads andy can you comment on that i mean yes it's true uh i yeah yes i mean you have different needs you have different storage requirements for those two different types of uh workflows and i mean it you can almost apply some of this to your primary versus secondary storage approaches where the the transactional workflows and the the type of things that you need to be able to have at hand are going

to be considered primary storage and the analytics and that type of thing would would potentially be something that you could do in your secondary storage um you know it this is this actually you know plays into the idea of what we did with at data gravity where there was a there was a front end node that actually served up your storage and there was a back end node that was the the secondary node on the storage but that was the one that was actually doing analytics

while the front end node was serving the storage and yes it's you you need to approach these two different types of data entirely different ways um yeah yeah i would contend that i would contend that uh in the old days yeah that was also that was very true but in these days and ages uh you can have transactional data sitting on uh things like object storage i mean obviously they're buffered and stuff like that but you know it doesn't doesn't matter as much anymore where the

data resides as long as you have sufficient quote-unquote middleware to deal with it you know well yeah i mean that's i i agree with you entirely there and you know my take my take on uh you know object storage is it's basically a file system without a directory structure yeah more or less although it's it's immutable that's the other side yeah immutable is it well i mean everything is more let's say blended in the past i mean if you think about in the rds era you

had to compete compartmentalize every single workload application because you wanted to you know take advantage of the internal tracks for faster access okay just you know get that millisecond from the from the from the fast tracking but in in other case short storage sorry but in other cases you wanted also to use the the hard drive for you know archiving so you use the second part of the disk to to put the data that you never access and so that was a lot of work behind

you know a single system and now you have flash memory okay so all primary storage systems and again primary in the sense of flash arrays with block and access and file access they are all flash even object storage is all flash now so you through the largest the ones that are fast yeah yeah yeah now that the fast object stores are all flash so yeah you don't really care about you know the type of workloads to some extent okay so i'm oversimplifying but actually you can

have many more workloads in parallel they don't really mess up with each other and again if you have enough middleware you you can optimize all the rights and all the all the uh all the caching that you need so it's also true that you know many databases now use object storage in the can maybe not for the transactional part or the most active part but parts of the database are stored in in the object store and excellent yeah absolutely and object store are starting to access

databases you think about s3 select s3 select you you put a csv file a as an object and then you can query the object i mean it was impossible a few years back but now it's something that and you are fluent some of the application intelligence from from that from there i think that go ahead and i i was going to say i think to further expand the the comment i made to start off here um with the transactional data you also need to think about

um sequencing and access and you know how to handle multiple access where quite often with analytics it's something that's a single access uh you know there's only one thing accessing it so you don't actually need to have any way of locking the data or making sure that things happen in appropriate sequence as opposed to analytics where lots of times you actually want the analytics engine to be very close to the data so you want to have very fast access to it and you know you might

not actually but analytics started out being a distributed hadoop cluster with you know triple redundancy of the data and it was all it was all distributed across you know dozens and dozens of servers so it's not a single use environment uh it may seem that way at the highest level of this structure but you know if you look at data lakes today they are you know they're multi-server clusters with multiples of storage i i'm not yeah i'm not disagreeing that they're the data lakes are very

large it's just you don't i don't think that you necessarily need the transactional consistency that you do with front-end primary data well it's becoming more and more i mean the other i mean so think about object stores again a few years back they were all eventually consistent now many of them the majority now is strongly consistent because the type of interaction that we do on these objects is way more frequent than in the past so it's almost transactional level yeah yeah yeah i mean for some

workloads it's critical to have to have this kind i mean there are several updates on the same object in uh in just a few seconds and you want to be sure to read you know what you wrote that that's that's the thing right and definitions where what does analytics mean right so self-service bi against data lakes and transactional databases is just a bunch of people connecting to the data and doing their stuff with it and then a well-managed pre-configured data warehouse that just runs almost batch-like

loads on a batch like processing it's completely different being right then i mean they really are they're not transactional and typically we don't update data warehouses we just append to them or delete and batch and re-upload and those sort of things so it is kind of a trick question because just like databases aren't one thing uh analytics aren't just one thing either right but do are there different uh i know andy you agreed that there are different storage requirements but does it get down to the

level of whether or not it uses block or object storage approach to it transactional versus analytical again it's just whatever you want your middleware to do exactly and it's i don't you know at the end of the day i don't really care how you store your bits and then you know everybody has their own flavor and middleware that they actually want to use to to get to it and it's that's a matter of how you prefer to consume it nobody sits around saying i'm i'm concerned

about what blocks this stuff is stored in you know unless you're a storage engineer and this is the type of stuff that it's like nobody really gives a sh how they're under how this stuff is stored underneath uh caterer can um yeah consider that everything is changing and if you look at infrastructure the abstraction that we have now in the infrastructure think about devops teams i mean some of these guys don't really understand what's you know what they have underneath their application and and they start

using object storage as primary storage just because you know it's an api that they understand and they can use easily and it's portable and everything and so they start using it i mean it's not that file or object is better it's just the api is is more in their you know way of uh thinking doing business developing software yeah yeah yeah exactly yeah and in spite of them being both developers and operations people who are supposed to understand both how computers work and how they use

them yeah yeah experience most of the devops team understand storage in storage classes right kubernetes guys have fast faster fastest and they think okay faster it's sold not really but you know well now you're you're preaching to me as the choir because of course these people also don't care about my responsibilities which is making sure the data is the right quality to meet the business needs because you're right they focus on performance and and performance for delivering code and performance of an actual system but let's

not care about whether or not this karen lopez is the same as that other karen lopez that's my normal rant there so it's it's the separation of duties that are important so we can have specialists but also that we have to come together and i can ask questions like does it make a difference between object and block that's not something i have to worry about in my day-to-day life but i should understand it more yeah you know it's you know their job is to make sure

the code is right they don't really care if the data is right underneath it that's true they care but just not enough they're never going to love their data the way data check loves their data um right so one of the things in the interview that keith did there was a lot of talk about high performance computing and how that was different and how it's the same i know andy you and i have talked in previous things about special storage needs based on different types of

high performance computing did you want to revisit that just a little bit about how storage is impacted within the different types of high performance like ai versus ml versus compute parallel compute those sort of things yeah to some extent it's i don't recall exactly what we were discussing but i mean the high the idea of high performance storage or high performance computing storage has kind of morphed over the years as well where high performance compute used to be like the world of researchers the world where

we were trying to do you know a zillion different computations to figure out you know like where where a proton is going to end up in your high speed accelerator or whatever uh and you know that type of thing is what high performance compute was largely used for a while ago and today you you get into the ai and machine learning world that is where an awful lot of high performance compute type resources are needed and uh the the idea is that you know in research

you need to be able to pick through an incredibly large set of data to be able to you know do the calculations you need to be able to find things and with machine learning and some forms of ai today you need to be able to go through an extremely large set of data to get inferences from it and figure out the information you needed to do and i you know i'm i remember we discussed high performance computing at one of the field day events but i'm

i'm struggling to remember exactly what we discussed we were talking about ai but it was good ray what do you think about that well i so machine learning is is effectively these days it's it's consuming awful lots of sequential data whether it's visual audio or textual or whatever but it's it's it's consuming this data it's consuming it lots of times and during the training activity you may be consuming this stuff you know a thousand times or 64 times depending on what you want to do uh

but most of that data they're consuming they effectively bring it into storage or bring it into memory rather and they're they're consuming it from a memory type of structure so or they're bringing in batches of it into memory and consuming it that way the high performance guys have always been you know [Music] massive levels of file i think uh enrico mentioned earlier luster ceph befs spectrum scale there's probably a half a dozen other ones that i don't even i don't even remember but they had massive

amounts of effectively filed data that that was petabytes or exabytes or yada bytes of store of data that they needed to go through and run some serious fortran against right serious scientific computing against you know you could argue that machine learning is is is a is a follow-on to that sort of thing it's a lot of mathematics differentiation and and that sort of stuff done on uh multi multi-core devices whether they're gpus or cpus or you know massive hpc clusters i don't know enrico you had

a pretty good discussion there in your email earlier today well in general i think the hpc in general is uh uh first of all the architecture are always very very balanced it's not that you have you know a storage that is particularly uh more powerful than the rest of the infrastructure so you see every time that they announce a new supercomputer the best supercomputer you also see a new uh a new architecture attached to it so and even the last supercomputer that was i think in

the us is now the fastest in the world they they spent a lot of time talking about how the storage infrastructure works and again it's not only a piece it's it's very complicated it's files there is a block there is you know tons of components so and uh it's layered architecture but the problem is you know these guys work so hpc is something that is a little really different from what we are in enterprise even the largest enterprise usually doesn't have the the needs of of

these guys yeah and they are for you know massive optimization even zero uh zero something means a lot for them and uh right and and they and they use this massive architecture hyperscalable yes they use fortran but they have also 10 000 nodes i mean yeah yeah yeah that's also the the networking is very very specialized as well it's not it's not your off the shelf infiniband or ethernet or something like that it's always extreme and so even if they talk files they don't talk 99

of the file about nfs or smb they have a client installed on so this file system and uh deposits client and of course you know they are both you know very advanced but also very uh very conservative because you know uh it's it's not easy to design this kind of infrastructure i mean uh and excellent yeah i think keith actually summed it up best a week or two ago when he he made the comment that it's like map is not a new function call it's you

know these these programs have actually been mapping files into memory for decades now and that's that's typically the way that an awful lot of things these things are going to operate on the underlying stored data no matter what they're they're going to map it into memory and they're going to treat it as if it's just a memory object um you know it's one of the things that i've actually been um discussing elsewhere is uh electronic design automation and this is another case where this type of

uh you know hpc function comes in you know extremely handy the idea that you actually need to run a zillion simulations on what's going on inside your field pen to actually get it to work the way you want it to and you know this is another application for uh like activities yeah i would like to make a distinction between commercial hpc and uh and you know academic hpc i mean see even if a eda is is really you know uh 100 an hpc workload usually is

run by you know uh component manufacturers and and they tend to save money okay so you can find a lot of nfs a lot of standard component and many of these workloads are moving to the cloud because they don't need to run these workloads all the time they don't really care about the infrastructure anymore so i actually they care to save as much money as they can so they move to the cloud because they can turn it off and on when when it's necessary okay and

the data is still there because the dollar per gigabyte in the cloud is very low it's when you access and move it that is expensive so the hpc in the academia is different because they make huge investments and they need to you know every penny has to count and and so it's everything is hyper optimized some of this stuff is customized uh you know and and in academia really really really opinionated like unlike us no opinions here an enterprise you you can find you know somebody

in the enterprise that listens and and at a certain point they make a decision that is more a business decision than a technical decision if you talk in academia where you have all these guys that study every day you know computing science uh computer science et cetera they they have very strong ideas sometimes they're wrong but most of the time are really as most opinions are yeah but but again it's very difficult to to force something different of you if they if it doesn't come from

us from there yeah something like that yeah so i think this is a good thing and maybe just the reflection is is that like you said research organizations including universities you know they invest heavily in their hpc because they can always make use of that time whereas enterprises just sort of need to rent hpc type services so as a service depends on the enterprise right i mean to a large extent yeah yeah to the most part yes i agree but you know some of these enterprises

are doing it more often than not oh yeah i agree with ray on that case it's i can actually think of lots of times where both both enterprise and government would have need to actually have their own resources yeah just today noah the weather people right just announced they did their press release on two new super computers that are being deployed their general dynamics delivered these two cray super computers that puts them now i'm gonna i've forgotten already the metrics on it because i just read

it but it's three times more performant than what they've had in the past so definitely that organization which i also consider like an operational and research organization for sure but if when i say enterprise a lot of times i'm talking about you know retailers insurance companies um financial companies banks stuff like that yeah that maybe are using hpc type technologies you know mostly for enhancing data that they have or calculating models i think there's an awful lot of modeling that goes on that way but i

also think that both in electronic and physical design corporations there's an awful lot of need for high density computing and high density storage and i don't think that those companies are necessarily you know just renting it yeah i agree i agree i mean basically the big answer is it depends right so the side track i wanted to make it since two of you mentioned it in the interview the the researcher indicate or the the person from the university indicated that fortran is the best language for

doing high performance computing now my take on it might be different than yours do really short very quickly what do you think about that statement because i thought it was a great statement python why would you use fortran anymore python's the answer the problem with fortran it's been around for 50 years or 60 years now and so every every one of these applications that they've developed over the last half century is written in fortran okay but if you're gonna do something today it would be python

i'm sorry just an opinion my humble opinion i don't know when i worked on a fortran project went way back when i was in university more than three decades ago i've done more fortran than i care to admit karen but i thought it was an old language then it was an old language then and it's my i mean yes there's an awful lot of stuff that goes on in python these days um but my continued take on this is that if you're actually going to take

advantage of your high performance compute resources and the storage that's underlying it you really need to be running a compiled language and that's when one of the places where fortran comes in that would be places where c would come in as well um i i have not been impressed with any of the python compilers i've seen out there but i really haven't been staying on top of it either python is largely an interpreted language and if you're going to actually spend the money to do high

performance computing do it in a language that actually is suitable for doing high performance programming if if you look at any of the machine learning activities going on and nvidia's super clusters and all those it's all python yeah it's super computing the libraries are not in python no no absolutely absolutely but you know the functionality that that's that sets it up yeah they're larger they're logically but when you when you call that the specific function uh is common not just because because we are talking about

a university and probably somebody taught them for none when they started you know their first compute science uh course like i mean fortunately they they they didn't learn java or pascal but no so this is great but i want to make sure we have time like one of the other things that we're supposed to be talking about besides the fact that someone believes all storage will be object in the future kind of embedded in some of this discussion is do you think everything everything we do

is going to become cloud-native and cloud-like architectures as you know maybe not in my lifetime but in some people's lifetimes people that are working in it right now you have to you have to define your terms here cloud native means something a cloud op you know cloud fun you know running on the cloud is another thing i think right now i i think you could state that you know if you have to scale if you have to distribute your workload i think cloud native functionality makes

a lot of sense there but not every workload needs to scale from one server to 10 000 in a matter of seconds you know if you're if you're an organization that has the scale requirements i think cloud native makes a lot of sense yeah and anybody else uh honestly so and again everything is developed in layer on top of layer on top of layers if we dig in still in a bank today we find that there is a huge mainframe behind a lot of microservices a

lot of other stuff so uh i mean when the last mind frame of this world will die then we can think about everything we'll you know we'll use the next technology but i don't think that mainframe will never die yeah it can technologies that we know even tape will never die will never die why would it die um ibm just announced today their development of uh their mainframe their zeta zos they said os on a vm just for dev test so like an emulator um so

that people could do that because mainframe is not going to die yeah it's you know cloud native or not it's i i think i think people don't really care they they want the consumption model where they can get the resources they need when they need them and it's whether you call that cloud native or whether you call it you know consumption driven um yeah i think you're confusing the terms here andy i think i think you know cloud native uh means really containers to a large

extent cloud native computing foundation is all about kubernetes containers and how those operate but if you look at the cloud a lot of the functionality running on the cloud is not cloud native quite frankly i mean yeah the hyperscalers underneath it probably all cloud native but you know i split up a couple of servers to do something i'm not necessarily using containers or or or kubernetes which are in fact two different things okay so nobody give me the glossary of what cloud native meant in this

i i just wanted to make sure you're aware okay no no but but i mean that if if that's what we mean by cloud native then no it's like no it did um i i think that you know people will consume resources the way that they want to consume resources and they will make themselves they will make it available to themselves that way and that's i i you know i i don't think the vms are going anywhere for as long as mainframes are not going anywhere

exactly come back in 15 years and vms will be the old way of doing things so let's hope so we've got like one minute left so what i'd like you guys to tell first i want to thank you all for having this debate discussion i i want you know me argument argument i want to continue it on social media as well um but i want to ask each of you to tell me where can we find you if we want to argue with you on social

media enrico so you can find me on practically every social media but mostly on twitter and lick it in and to find my profile just google me because there are not many erica similarity on working in ap yeah right um mostly on i you know social media i've been kind of uh going off that that tangent i'm on twitter at rayleigh casey i am i blog at rayonstorage.com i'm a podcaster in graybirdsonstorage.com so any of those you can comment uh at will again and yes you

can you can certainly google me and there's probably four or five rayleigh cases but not many in storage so we should just email you our debates yes if you wish and i could happy to send you that if you will and uh you can find me on twitter at andy banta and there are more multiple andy bandas on twitter but there's only one that has the handle at andy banta there you go that's how twitter works so you can find me as data chick because with

a name like lopez i have to have a different name so thank you guys so much again and i want to thank cto advisor and keith for inviting us to argue in public thank you