 Got we got sound yet sort of yes. Okay, sweet. Okay. I hope I hope nobody else is feeling as dusty as I am this morning But never mind moving on So quick a quick intro to what we've been doing with open stack and other things at the New Zealand a science infrastructure So I'm I'm the dusty guy on the left there We've got Mikael here from Stack HPC and well Sean's nowhere to be seen yet, but we'll find him later Okay All right, so just a quick intro to Nessie so We we support a bunch of different science in New Zealand where a national collaboration of the University of Auckland Neewa University of Otago and Minaki Fenua landcare research So we're a team that's spread throughout New Zealand. You can see we've got staff around four different sites Supporting a bunch of different research and science across New Zealand. So mainly mainly in in the research space But we're also working with the crown research institutes in New Zealand And so they're doing also translational research workers as well so we we come from a high-performance computing heritage and The this is our our primary data center and and the main machines that we have in it at the moment You've got on the left hand side there Maui. It's an XC50 crane machine and on on well, sorry on the right on the left is mahuika, which is our capacity cluster system and And that's really you know, so that's that's our bread and butter The storage in the center we have So high-performance computing and data analytics service We do a lot of training as well around the sector There's data transfer and sharing as part of that platform and we run a consultancy service as well So we have computational scientists working with researchers on their codes So what are we doing with OpenStack? And why are we here talking about HPC? Well That those platforms that I showed you there They they originally had some OpenStack built into them but it was very tightly coupled and didn't quite allow us to do the things that we were hoping to do there which was to bring in new styles of access to be able to give users their own environments provide multi-tenancy at the platform layer and so we run we run a thing called our platform our national platforms framework as As our investment mechanism and and how we decide, you know, what new systems that we should be looking at services that we might bring in And so last year we put together a case around both scaling up some of our existing infrastructure and and bringing in more GPU capacity and those sorts of things but also around Fundamentally building a new platform to address these new styles of access provide increased user user user interactivity That get that data and computer isolation with a more flexible design than traditional HPC and mainly at least for the first year or so supporting our learning in that space and operating such a platform So that was enabled by a pretty small investment From our platform access fund which is so we take subscriptions for access to some of the HPC services And so we've reinvested that to to build this platform initially so this this is a Quick look at how we're sort of viewing the services that we might be offering out of this platform And you can see that this is more than just open stack So open stack is clearly it's key in there at the platform layer and providing that multi-tenancy But we're thinking about this as a broader ecosystem of of services that we might be offering out and also Ways of bringing tenants into the environment as well so you can see you know going up from campus network integration and Data management and some of these things that aren't part of the open infrastructure stable But are of course very relevant in in the research and in the HPC domain so When when we when we propose that investment We also had a few key use cases in mind that we're going to use as pilots to inform our learning through that so couple of those Here we've got the of course. There's there's some internal service development going on just actually giving giving our DevOps people our support people A platform that they can use to to build things without Requiring route access on on the HPC system those age-old problems But also so we've got a partnership with genomics Artira, which is a another research platform in New Zealand funded by the government to build a genomic data repository and another pathfinder project looking at using clip genomic genomics in clinical medicine and using creating an environment with trusted workflows and Limited access to patient data So the the Artira or genomic data repository That that's been quite a success story. So that's in production now That is a tongue of species Database so tongue of species are or taonga is a treasure for in in Maori culture And so that's been a co-design process with genomics Artira and Iwi To build this database and govern access to it So it's high-quality genomes and researchers can apply for access to use that for particular use cases It's only a small database, but for us technology wise That's a it's a cloud native microservices architecture. It's built on a product called gen3 and so We've had people working on that and then deploying that onto a Magnum on OpenStack. So one of the other pilot use cases is research software engineers, so we're sort of not looking at Going out direct to all researchers and providing them access straight on to an infrastructure as a service cloud But thinking about the research software engineer and I guess, you know, the the DevOps within a research group and what they need And so at the moment we're working on Identifying those user journeys and and some of the value-added services that we can provide for that class of users There's also a partnership that we have with a group at University of Auckland the strong AI lab sale to bring some Large GPU capacity into this environment. So I showed you that tenancy layer cake before so one of the key Features of the platform is multi-tenant bare metal and so that enables us to Have groups actually bring their own hardware in and and use the platform to manage access to it So another big use case is a is a recent partnership with one of the crown research institutes in New Zealand ag research So they had a small internal HPC capability and a fairly large but Unwieldy data environment for campuses with NAS boxes spread everywhere backups going to and from and that was essentially getting out of hand for them. So they went to market and Looking looking for a solution to this eventually partnered with HPE to provide them with hardware For a new environment But then realized that they they didn't quite have the capability to be operating it themselves And and to be realizing the evaluated services on top of it. So they came and partnered Looked to partner with us and that this was at the time when we were designing the flexible HPC platform so we we took the initial design and then partnered with HPE to Fit that into the flexible HPC platform. So you can see the diagram here the blue parts where we're bringing Aspects of their platform into the flexible HPC service. So the the Tamaki data center piece over here on the left is an open-stack environment over here, we've got an object store at a different site and Then there's a there's a bunch of other pieces that are outside of the open-stack environment but part of their tendency So a big data infrastructure a GPFS file system DMF a tape library zero watt storage some of these things So that's a that's a pretty deep partnership and actually that that investment and and their their environment actually Dwarfed Nessie's initial investment into into this platform But I guess, you know, we must we were doing something interesting and so, you know, they wanted to be on it. So that's great It's a quick look at the back at that tendency diagram looking at the ag research tendency So you can see here we've got some some of the shared services in in blue that they're Making use of and then some you know some tenant specific things in green All right, so moving on from the use cases From an infrastructure perspective the platform like I said multi tenants bare metal infrastructure supporting high performance workloads cloud and bare metal provisioning Onto a common network What else we got? one of the interesting things that we've done here is on the slide is Partner with the National Research and Education Network in New Zealand, which is called Rianne's To have direct connection onto their network. So we're using a Layer three fabric within the data center network for this platform And we're actually peering directly from our core switches in that fabric Onto Rianne's And so one of the things that that's enabling us to do is have we have a public network, of course, and that's within our VRF zero on on that in that networking environment But then ag research for example have a VRF as well So they were actually able to bring provider networks into that environment which are within their private WAN on Rianne's so that's quite a powerful capability And you know with with no VPN overheads or anything like that So capacities and so on we've got so like I said you can the dot points over here our initial investments quite small It's only about a thousand vCPUs 30 terabytes of flash storage half petabyte of object And a few a few GPUs And of course the the 100 gig fabric a Few you know pictures of some of the technologies and so on that we're using there If you were at the scientific sig yesterday, I mentioned some work We were doing around identity using key cloak and and free IPA so I'm gonna pass over to Macal in a moment, but I just want to say we've been working on implementing this platform with stack HPC And so one of the things that we see is key here is Building an open infrastructure We we're not particularly wedded to only doing on-premise cloud, but we do want solutions that are fundamentally not locked in And so the partnership with stack HPC here has been pretty key because they work upstream first And so there's been a few things along the way here, which Macal can tell you about Where you know, we've had to fix a few things Thanks Yeah, so from a technical perspective one of the most important things was to manage the Networking fabric with BGP VPN using Kyobies physical switches integration. So we've prepared Ansible templates for doing that and it works quite nice We're also doing Changes in networking generic switch Newton driver to support Bermuda instances with M lag bonds and and trunks to support multiple villains over that link We're using Oven in this deployment So we have Enabled open V-switch hardware offloads with Melanox connect x6 cards So connect x6 is important in that because we can offload the connection Tracking flows so that means security groups and so on so we can use security groups on the offloaded villain networks Using SRAOV as well And our DMA Using Melanox ASAP square technology. So for currently we're using Newton DHCB agent from the MO2OV as stack For external bare metal ports because that functionality isn't it's still not there in Newton, but it's in progress So we should be moving to native Oven in probably couple of weeks time Yeah, so We've also contributed to co-answerable like FQDN based HAProxy setup because The default in co-answerable is just exposing frontends on different ports and we wanted that to be like Better for access and the users because some ports some sometimes are filtered and so on so so it's FQDN based everything is on the SSL port exposed to the internet Yeah, and also we've worked on Magnum To enable all of those workloads, so we've enabled SMT on Federa Koro S The patch is in progress upstream currently being reviewed I've also added Octavia when the provider support in Magnum and in the Kubernetes OpenStack provider Repository managed by Kubernetes SIG Yeah, I think we are over time Yeah, we fixed bot-to-pot networking in the MO2Oven. There's a bug So it's currently a workaround Newton guys are working on fixing that permanently and We also use many less FFS shares on those Magnum clusters We still need to propose the patches upstream Thanks we go All right, so that that's it Also, we're hiring So, you know if you feel like moving to New Zealand talk to me So any questions? Yeah So You must oh good, and then we better get off. Thanks