 Yeah, I think let's get started Alex. Okay, let's see this So in terms of some general housekeeping from the From the previous call the end user survey was was sent to Cheryl to be to be forwarded on to the end user forum We have an end user forum meeting Booked in for the night of July Which is after the keep going in China We will We had We had the CNCF sick charter forwarded on to the to the talk for discussion and My understanding is that they that the talk we're going to to review with a look on 15 on this prior to coupon Is that your impression to Clinton? Sorry, I lost my unmute button. Yes, I believe that it's all kind of happening And they yes that that is the plan. I'm not aware of any obstacles Excellent, okay, so So in terms of in terms of next steps We were discussing options about putting together some use cases Using cloud data storage We obviously We obviously wanted to get feedback from the end user community, which will be getting through the survey But in the meantime I just was to pull this group to see if there were any specific use cases That you fell sort of out of that you have the resources or or or some some basic information to To be able to move forward as well. So we could have a list of those for consideration Do you know when we expect to get some results from the story I put the deadline for the 16th so that we can get the we can get the summary of the results before keep going Oh, okay, it's good So I'll I'll kick off. I'll kick off This was it so the idea for the use case is myself. So I had an idea that you know even the Priorities of the end users assigned I think it would be useful to cover a Use case document for some common options for for volumes. So sort of covering the ephemeral volumes the the local volume support that's been recently introduced in Kubernetes and persistent storage from When using a storage provider to cover, you know, some some different use cases or different Applications that kind of fit to those three types of volumes and because this is quite possibly a Fairly common question we get asked at storage of us, but I'm open to sort of hear Comments from from anybody else on the team was to whether they think this is a good idea or whether they would rather we did something else I was just wondering. I'm sorry carry on Which is saying that The block Rob lock sport is available now. So maybe want to know some use case for that okay, yeah the the question I had was is information not perhaps already obviously publicly available I would imagine with the release of A local volume for example, there's presumably a blog post that explains what the use cases are for them or is that not the case? Yeah, there was a blog post with the with the feature release. I was I was thinking I Was thinking more something along the lines of you know, this is this is you know This is these are some examples of what you could use a federal storage for these are some examples for The local or the Roblox store and these are some examples of when you do when you want, you know an external persistent system or service To kind of help people with just general guidance on these things Again, I mean, I haven't actually been following the blog post associated with the releases of those features But I would have imagined that was already covered if it's not already covered and then yeah, probably falls upon us to clarify that Okay It would seem like a shortcoming in Okay, so I think there are some blog posts have more Use case cover like like for example the local voting I think there is a pretty good block covering use case actually from I think Uber's using that or something But for the Roblox for the blog post doesn't really cover Clear use case. So I think that one maybe we can See if we can you know add more to that At the risk of getting sort of overly philosophical. I wonder if we shouldn't I mean if projects are releasing features without a clear explanation of What the feature is intended for I did say that I'm just saying that I Was more thinking do we want to look at it, you know, if user, you know, actual users using that for something, right? Yeah, rather than just say, okay, this can be used for some okay No, that makes sense and actually I had a similar thought stroke comment when Alex was Soliciting agenda items yesterday. I think this word use cases kind of become Overloaded in the sense that and maybe we need to come up with a new with different words, but in my mind, there's like Specific user has specific, you know Project where they did something and other people might be interested in understanding like what exactly they did and why they did it And whether it succeeded etc Now, I don't know if we call that thing a use case or something else My sense is that we probably need to call that something else and then there is, you know In a more generic sense like what are the categories of use case for this feature? What kinds of things what are the properties that you're looking for, you know a more generic kind of description of something and That also I think currently gets referred to as a use case and maybe that's Right, so I think I think that's a good distinction. I don't know what to call The latter one where we're talking about something for from an an actual say and user. This is how we did something I was I was thinking about this specifically because it's something with It's something I've certainly come across and some of the the people I needed meetups have come across Where different classes of volumes, for example Are used incorrectly for different use cases. So for example A classic example of this is the official at CD operator Actually uses a thermal storage So if all the nodes have a power supply failure at the same time You will actually just lose all of your data because that's the way the operators designed and I think it would kind of it would benefit a lot of people to avoid these sort of stomach blocks and to sort of adopt some of these Client into technologies quicker If we just said These are the types of volumes that you can have just because you're you have a PV from from from a CSI driver or a native driver Doesn't mean it's going to be persistent doesn't mean it is going to fail over and all these sorts of things These are some of the the categories and therefore, you know, use them accordingly kind of thing and I was I was thinking about it in terms of You know short circuit in some of the pain points that that users are finding when they're using the official repos Yeah, okay that that that makes sense and I think that that kind of stuff falls very squarely in in our set of responsibilities in the sense that we You know across multiple projects and across multiple Yeah Things we need to sort of provide a comprehensive understanding to the end users However when it comes to a specific feature release for example local ephemeral volumes for Kubernetes or whatever one would hope that those are described by the Kubernetes project and and in cases where they're not where we Don't think they are sufficiently described. We should you know help them Fix that and and not do it again in the future. Otherwise, we're going to end up Documenting all of the CNCF projects, you know, which I don't think is our job and I don't think we want to be doing No, no, I can be keeping I'm not suggest and just to clarify. I'm not suggesting that we do that Provides kind of more about summary Compares and kind of docs so that people can choose to write a bunch of rather than sort of sort of Yeah, it's not the future releases Actually, just as a slight diversion. I don't know if I understood you correctly, but if I did then you were saying that that etcd using local ephemeral storage is a bad idea and I Think that's not true So I just want to make sure we haven't stood each other and and yeah, whether there was some confusion. Yeah, so so so This this came up for example The other day when I was discussing this at the meetup The the actual helm charts today installs It's kind of runs up its cd on ephemeral storage, which basically means when the pod dies that Instance of scd goes away now You know if you're running as in as many users do it say just three notes or something like that It's quite easy to get into a situation where you lose quorum or you lose all of the three pods at the same time And that basically means that there's no data to recover. There's no snapshots. There's no backups. There's it's just gone And so, you know, there are certainly arguments for whether you you know You want to take snapshots if you're using ephemeral or whether you want to use something else like for example the local volumes Because at least then you can restart the pod and recover the data. Oh I understand so your your distinction was between ephemeral versus local rather than Local versus remote Correct. Yes. Oh, I understand. Sorry. I wasn't actually aware that there was a distinction. Okay, that makes sense I was just the reason I asked was was There is a perception that storing at cd data on remote disks is a good idea and and That it's a bad idea to use local volumes and I'm not sure that that does Sufficient just as to the debate about the two options I completely agree. Yeah, okay. No, that's good. And now I understand your point Cool, all right So so in that case I can I can potentially maybe sort of put a Put a document outline together to cover together this and then maybe we can work Collaborately to split up different sections of the document. Hopefully this can be a nice short document that we can turn around That we can turn around quickly. It's not going to be something as meaty as the white paper Yeah, sounds like a good idea. So just to be clear is your is your thinking to do it like a kind of a Q&A, you know Should I should I use these kind of blockstores or these kind of blockstores? You know bunch of questions and answers or are you thinking of it as something? Something like that with a paragraph on each type and a comparison table or something and that's it so that it's kind of like a two-page Reference article if you wish Okay for a very specific kind of pair of alternatives Yeah Okay, that makes sense. Yeah, because the the white paper tended to sort of try and sketch all the possible alternatives and and pros and cons of each which tends to be a much more Difficult thing to give a clear answer to so it tends to be begging in a lot of areas by So you okay, that makes sense All right, so so we can get cut off the the Are there are there any? sort of other areas that the team feels strongly about or would like to to Think about or to try drafting Alex on the same lines that you mentioned about the local PV and decimeral storage does it also Make sense to write something on hyperconverged storage versus remote storage When it could be a good option to use one way or the other Yes, that is That is a good idea actually It is something we touch on in the white paper and and we have asked the End users about that in in the in the survey something we touch on there. So Yeah, that might that could be another good option. That's good idea And we can probably pull in the responses from the end users survey and feed into this Thank you kind of an article or the Same same format that you are using for a Yep, absolutely. I happen to to read that section of the paper recently and It's fairly brief, but I guess what I'm wondering is how much more there is to say My recollection of what the paper says is essentially The pros are that you have, you know, a uniform commodity hardware with a converged approach And it's sort of more flexible, but on the other hand your capacity provisioning and performance impact between your compute loads and your Storage loads gets mixed up So which I think was a very good summary of the My understanding of the pros and cons. So I guess what I'm wondering is what more do we think we need to add to that? It's not obvious to me that there's anything that's not answered by those questions, but maybe you have other ideas As more looking from the end user use cases, what are the scenarios where they are going for one option for the other? Is it more like an edge use case where they don't have Access to the remote storage or is it like an NBME? Kind of a devices which are available on the modes No, just trying to So I Kind of understand what printing is saying. I think we covered the basics in the in the white paper quite well What would be useful is to try and capture Some of the some of the motivations from From the end users and for some, you know specific difference Use case examples, I guess so so just by by example by word of example, sort of I I have come across different users have chosen different apologies for reasons completely separate to You know some of the technical aspects that we govern the white paper. So for example And Commercial reasons are one thing You know where that where perhaps they're looking for say optimized storage notes versus optimized compute notes But other reasons which are fairly more common are actually Operational and and sort of day two operations type thing. So for example They might have a higher rate of change on their other compute notes, especially if they're running with managed services in the cooking in cloud environments Then they may want on their storage notes, for example, I Don't know whether that's actually merits an entire paper or an entire documents. I Don't know open to date I mean, I guess you can we can we can put us on our we can put on our list And use the information from the survey to confirm one way or the other Yeah, that makes a lot of sense to me anything else Well, the other one is the the roblox that I mentioned as I think it would be good to see I know it's still relatively new. I'd like to see Who is using it and You know, if they are they think this is something they need That maybe helps them improve performance versus the the file system mode But this I think I need some user input Yeah Because we know in general what is why that is needed, but I'd like to see Someone is really using it and the same This really helps us Yeah So I don't think we have that specific question in the question. Now. Maybe this could be like a follow-up It could be do you know what actually Every time I sort of read about the roblox support one of the obvious things that sort of occurs to me is that this that's actually a neat feature for Storage systems to actually use right It's usually for some of the database applications like SAP actually they want to have the roblox or that's the We do that's one of the use cases, but I still won't you know Let to know who is actually using it since it's still relatively new right Remember whether we had that in the survey you do you think it's not I don't think it's I don't think that this is that We just say the general, you know blocks are you using blocks through a system or not because this is very specific This is like a one mode in Kubernetes Right so the for the for the PVC, right? So so we don't have something that's specific in our survey yet or our service kind of a general Well, it's certainly something we can bring up When we when we speak to the users Yeah, that's a good idea. So that is after they have done the survey and the after the keep calm. Yeah Good Yeah, one one last item Just sort of brain dumping is that as a sig as we transition into a sig one of our responsibilities is to Well, the two things that we have not historically done and that we will be responsible for in the future our project health So all the CNCF storage projects Just checking in with them making sure that they're happy and Healthy and and whether there's anything that needs to be brought to the attention of the TOC and then the other area is Identifying what we believe to be gaps in the storage Portfolio of of the CNCF So I don't think either of them are like super burning urgent like need to be done in the next couple of weeks But but we do need to make sure those get on our agenda and we have a plan going forward for how we can address them That was good. Yes degrees So, yeah, maybe you can just put that on the sort of agenda for the future Etc. Yeah, definitely Okay One other one other project or initiative That Like so it might be helpful, but I'm I am looking for guidance here because I don't understand that this could be potentially controversial so I'd be interested in putting together some Guidance for how we Actually measure some of the different attributes that we defined in the light paper So for example, you know something like performance might be a good place to start with something that's Quantifiable, but as we all know benchmarking is notoriously complex and subjective But I was kind of wondering whether it would be Useful to put together, you know something like a sample set of either Benchmarks or tooling or or even just maybe a standardized sort of automation framework for some of this stuff Such that end users can easily Use that framework as a as a starting point to stay for when they're comparing Different cloud providers or or or different Storage systems Within their environment and you know the idea being that we start a performance, but perhaps Move it to other areas where we have For example things like availability or something like that. So for example I know the opening the skies have have a Framework which which perhaps could fit the bill in terms of things sort of chaos monkey type testing with these type of storage systems But I was wondering if we could kind of maybe put an effort to sort of Standardize or bring together a number of people from the community to kind of put a I Think I think the general principle is very good I think having a place where users can go and Find a set of tools that we believe to be useful for this kind of stuff and maybe even some Performance results that we've done on you know either CNCF projects or general other projects other I mean in the extreme possibly even products although that gets Even more controversial My reservation is whether we have the bandwidth and the longevity to You know, I don't want us to build tools that we don't maintain over time So if we're gonna build tools we need to actually sign up for maintaining them properly over time And I don't think as a sig we have necessarily the infrastructure to do that properly And the same goes for performance, you know, if we publish performance figures for version X or product Y or project Z We we probably implicitly are sort of signing up to do that, you know periodically And then all the gaps and I'm a little nervous that we might be biting off more than we can chew there so so I I Completely agree and just to be clear. I'm not proposing that that's me the sort of published results because I agree with you they'll they'll get out of date Very quickly and it's apart from the fact that you know, we don't really want to get into Sort of a comparison of products ourselves. What I'm talking more about is this kind of providing Say for example, you know, there might be two or three tools like FIO and syringe and a couple of other things for example that measure performance for particular use cases But if we if we tie it together with Some simple automation framework with some sort of default known good values That make it easy for somebody to just say, okay, here's my storage glass. I'm gonna run up this this benchmark And I can do the same thing for some different storage glasses or some different cloud providers And it gives the users effectively capability to kind of try out this stuff themselves Then we kind of give An unbiased standardized way of people to compare numbers between different systems themselves And if they want to publish stuff, well, then that's kind of up to them to publish stuff But but we're not actually publishing it as a sake Yeah, that makes sense. I would propose that as a first step, we we just Survey the landscape of available tools and that can be used these kind of measurements and perhaps publish a white paper just saying, you know, these are the ones we Looked at these are the pros and cons of the various tools These ones measure these things and these other ones measure these other things and here's you know A couple of sample runs maybe of some of them just so that you know If somebody's wanting to do performance measurements on storage systems, they they can read this paper and get pointed in the right direction That that seems like a useful first step Again, the the idea of building automation implies that we maintain that stuff and keep it working and these things, you know, they Presumably that automation is going to rely on underlying tools which change over time and You know may or may not run on particular operating system platforms, etc And that's where I don't want us to like implicitly sign up to maintain a bunch of software that we don't actually have the people and with to do properly Let's in that context some of the users actually Gave links to some benchmarking tools to use maybe in the upcoming Meetup I can probably present some of those things and of them or how they use it Yeah, that would be great. I so I like the idea of starting small we can start small with With a survey of some of the tools and provide some So this guidance on which tools to use And then I guess depending on the uptake we get from the community we can We can make a call as to whether we want to take that further. I I completely sort of agree with you Quinton We should not and must not bite off more than we can chew But I kind of feel that with potentially dozens of Snorage vendors and companies who are sort of sponsoring this and could be interested in the SIG We might actually get We might actually get a bit of a community around this because I suspect that this is a common problem for end users and The sort of storage community alike. Oops. I think we last year Alex. Oh Can you hear me? Yes now we can. Oh, I'm sorry. No, I was I was just saying I think we may find that that's so there's a bit of a drive from both end users and The the vendors and the community around the SIG That that we may find is there's quite a bit of uptake in this So I like the idea of just starting small with the put, you know, a dock with a Landscape of the available tools and then kind of taking further based on uptake Yeah, I totally agree and I think the general principle of making real hard data easily accessible to Users is is very I strongly support that because I think as you point out, you know We wave our hands and we say that, you know, these kind of stores are faster than these kind of stores or more consistent or whatever But I don't think people appreciate quite how many orders of magnitude difference. They are between some of these things And when when we sort of make statements that when you use the wrong stack of storage Technologies or not necessarily the wrong one, but if you use a particular Combination of layers in your storage system This is the result in terms of performance or consistency or whatever the property might be and and it is like, you know X number of orders of magnitude different than than a different approach I don't think people always appreciate quite how big the differences are Yeah, cool, I'll stop by playing on that So So from from my parts that was That was all the things I had on the the agenda Does anybody have any other business? Not from my set. This is orders to Vegas. I was listening at the Colt and Probably didn't introduce myself. I'm Product manager at the EMC Oh, hello, welcome Hello, cool. Okay, so if if we've covered everything I think we get 20 minutes back on this Anybody has anything else they want to cover? No, sounds good. I assume someone's gonna write up some meeting notes. I think they're kind of empty at the moment Yes, get a note taker for each meeting in future Alright, thanks