Edge Clouds with OpenNebula
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
| Title |
| |
| Title of Series | ||
| Number of Parts | 490 | |
| Author | ||
| License | CC Attribution 2.0 Belgium: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor. | |
| Identifiers | 10.5446/46879 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
|
00:00
Graph (mathematics)Graph (mathematics)Physical systemContext awarenessComputer animation
00:23
InformationObject (grammar)Military operationGraph (mathematics)HorizonComputer programmingGraph (mathematics)Computer animation
00:54
Graph (mathematics)Demo (music)Virtual realityPhysical systemHybrid computerCloud computingSoftware frameworkDistribution (mathematics)Directed graphService (economics)SoftwareOpen setSoftware frameworkMereologyInterface (computing)Goodness of fitPower (physics)Graphics processing unitService (economics)Graph (mathematics)Virtual machinePhysical systemOpen sourceElectronic mailing listPerspective (visual)
02:30
Graph (mathematics)Web pageWeb pageGastropod shellMereologyPerformance appraisalIntegrated development environmentSoftware testing
03:08
Demo (music)Graph (mathematics)Extension (kinesiology)Open setGraph (mathematics)VideoconferencingHeat transferInformation privacyInformation securitySoftware maintenanceOverhead (computing)AerodynamicsUniform convergenceArchitectureDisintegrationDevice driverGraph (mathematics)MereologyFocus (optics)PreprocessorMultiplication signLimit (category theory)ImplementationComputational intelligenceData centerDifferent (Kate Ryan album)Natural numberSoftwareComputer hardwareOnline helpSoftware maintenanceVector potentialOverhead (computing)Form (programming)Dynamical systemInsertion lossPerspective (visual)Information privacyBuildingComputer animation
05:55
Asynchronous Transfer ModeFingerprintGraph (mathematics)Visualization (computer graphics)Cloud computingVirtualizationDistribution (mathematics)Graph (mathematics)ExistenceBuildingOperating systemService (economics)Physical system2 (number)Internet service providerMereologyOpen sourceComputer animation
07:36
Address spaceArtificial neural networkData storage deviceInternet service providerSimilarity (geometry)Limit (category theory)VirtualizationGame controllerAddress spaceComputer animation
08:19
Artificial neural networkLatent heatIntegrated development environmentVirtual LANInternet service providerData modelComputer networkOverlay-NetzIndependence (probability theory)Address spaceData managementDevice driverGraph (mathematics)Demo (music)Graph (mathematics)DisintegrationConfiguration spaceArchitectureCovering spaceProcess (computing)System administratorInternet service providerMereologyCASE <Informatik>Configuration spaceSystem administratorVirtual LANUnicastingverfahrenBookmark (World Wide Web)SoftwareComputer architectureDevice driverService (economics)IP addressSet (mathematics)Data managementNetzwerkdatenbanksystemSoftware frameworkINTEGRALOverlay-NetzGraph (mathematics)Address spaceVirtualizationRow (database)Computer hardwareIntegrated development environmentSingle-precision floating-point formatLatent heatLimit (category theory)Physical systemComputer animation
13:18
Graph (mathematics)Process (computing)Message passingVirtualizationIndependence (probability theory)Graph (mathematics)Computer animation
13:38
Data managementTemplate (C++)Configuration spaceSystem administratorGraph (mathematics)ArchitecturePartial derivativeArtificial neural networkState of matterDevice driverDisintegrationScaling (geometry)Library catalogLibrary catalogDevice driverComputer configurationInternet service providerData storage deviceScaling (geometry)SoftwareINTEGRALPower (physics)Arithmetic progressionGraph (mathematics)Computer architectureData managementVirtual machineCartesian coordinate system
15:22
Open setIntegrated development environmentGraph (mathematics)Single-precision floating-point formatGraph (mathematics)Projective planeSheaf (mathematics)Performance appraisalConfiguration spaceData center
16:13
Demo (music)Lambda calculusGreen's functionGrass (card game)CASE <Informatik>VideoconferencingGraph (mathematics)Sanitary sewerGame theoryServer (computing)Graph (mathematics)Office suiteUniform resource locatorMultiplication signCASE <Informatik>Video gameVirtual machineValidity (statistics)Computer animation
17:13
Demo (music)Lambda calculusGrass (card game)Green's functionCASE <Informatik>VideoconferencingSanitary sewerComputer animation
17:39
Default (computer science)Fluid staticsConfiguration spaceToken ringDemo (music)Physical systemComputer networkGraph (mathematics)Internet service providerComputer hardwareConfiguration spaceOpen setSoftwareDefault (computer science)Block (periodic table)Type theoryDevice driverAddress spaceData storage deviceMereologyIP addressDemo (music)NumberComputer animation
19:00
InformationMilitary operationObject (grammar)Bridging (networking)StatisticsSoftwareData storage deviceComputer animation
19:22
Bridging (networking)StatisticsInformationPhysical systemDefault (computer science)Data typeGraph (mathematics)Computer fileComputer-generated imageryElectronic mailing listVirtual machineMedical imagingLaptopComputer animation
19:47
Template (C++)Computer-generated imageryPhysical systemDefault (computer science)Data typeGraph (mathematics)StatisticsComputer fileGraph (mathematics)SoftwareTable (information)MereologyAliasingComputer animation
21:01
Virtual machineComputer animation
21:26
Virtual machineSoftwareVirtualizationMedical imagingComputer animation
22:21
Maxima and minimaLink (knot theory)Hill differential equationBootingAddress spaceComputer animation
22:44
Hill differential equationMenu (computing)Nim-SpielConvex hullLink (knot theory)Data typeStatisticsGraph (mathematics)Computer-generated imageryDefault (computer science)Computer filePhysical systemVirtualizationSource codeComputer animation
23:15
Data typeGraph (mathematics)StatisticsComputer-generated imageryDefault (computer science)Error messageTemplate (C++)Machine visionInformationObject (grammar)IP addressVirtualizationInternet service providerComputer animationSource code
23:51
Hill differential equationLink (knot theory)CASE <Informatik>PlanningComputer animation
24:24
FacebookOpen source
Transcript: English(auto-generated)
00:06
Hello, welcome everybody. My name is Vlastim Manohar. I work for the OpenEbola systems. And right now, we will speak a little about the edge clouds and how we understand the edge clouds in context
00:21
with the OpenEbola. I have to say that I won't do any real demo, but I'm starting just to show something. And I will describe it a little later, so just for the reference. So about the edge cloud, it's necessary to say that our work, our current work, and our future works
00:44
receive the funding from the EU Union from the Horizon 2020 research program. This talk is divided into mainly three parts. In general, what is OpenEbola, if you don't know.
01:01
Then, how we understand the edge clouds. And then, what do we do or what approach or what tools we provide to build the edge clouds in the OpenEbola. So just a quick poll, who knows what is OpenEbola? Just a quick, good. And who uses OpenEbola?
01:23
Great. So just to summarize, OpenEbola is a framework to build infrastructure as a service cloud. Mainly, we focus on the private cloud. We talk about virtual machines, powered by KVM or vCenter, and system containers powered by LXD.
01:40
We support various cloud deployment architectures, but mainly what we focus on is on-premises private cloud. Usually, OpenEbola is appreciated for being light and simple, extensible, easily upgradeable when compared to other systems. It supports various popular Linux distributions,
02:04
CentOS, Reo, Ubuntu, Debian, and so on. And it's fully open source under Apache license. And it's with us for some time already. This is how it looks like from the perspective of the graphical control interface.
02:21
A list of virtual machines at the top, a list of buttons you can control, reboot, and the power of distro and this kind of stuff. If you want to see all the features, there's a discover page which lists them. If you want to try the OpenEbola,
02:41
there's a really cool tool written in a shell called MiniOne which simply configures all the OpenEbola, like a front-end part and the hypervisor part, on a single selected note. Simply creates some kind of evaluation or testing or maybe development environment.
03:01
So it's really great and it takes just five minutes or something. If you're interested more, let me invite you to the OpenEbola conference, which is like every year. This year it's in Brussels, just a few months later.
03:25
Now let's focus on the Edge Cloud. What is Edge Cloud for us? For us, the Edge Clouds are the micro data centers which are with some kind of cloud-like capabilities
03:41
and are deployed very near to the end users or end devices they need to interact with. So the benefits are to provide lower latencies, provide new features which this lower latencies allows.
04:00
Probably some pre-processing can happen in the Edge Cloud so the data which are sent to the central cloud are much, much lower. Or there can be a need for it due to some security on privacy reasons. But everything is not just a green or good. There are some limitations.
04:21
Limitations comes with a limited offer over the hardware or software we can use in the Edge. And there are also some risks, potential data loss or maintenance overhead and so on. But from the perspective of the OpenEbola, Edge Clouds are very similar or should be very similar
04:41
to the on-premises cloud which is on the left side. So the main difference is we can expect there will be much more such smaller clouds. And they will have kind of dynamic nature. They can be created dynamically, destroyed at any time we need.
05:04
So they are kind of ephemeral. They are also restricted and kind of limited. From the implementation perspective, we talk about the infrastructure Edge. So that's the part of the Edge which is powerful enough to run like a more demanding computation and so on.
05:28
Our aim is to take the technology we are using for the on-premises cloud, all the KVMs, LXD, VX LANs, and all these building parts and just move it to the Edge Cloud.
05:43
Of course, with help of some specialized drivers and in a form which is like a design to run in the Edge Cloud. We very much rely on the existence of bare metal cloud.
06:04
So OpenEbola never installed the physical host operating system and these kind of things and it won't do. So we expect there is some service provider which manages the infrastructure and is able to give us or give users the resources
06:20
which can be used to build the Edge Cloud. The second important part is automation. Everything is automated. On bare metal cloud, we don't like a card that much. It's the provider's duty. But on the host level, everything is automated.
06:42
Hypervisor is installed, operating system is configured to run as part of virtualization cluster. And it's good to say that the only thing we do is just take all the great open source tools we have, distribution hypervisors and so on,
07:02
and put it together to build open source Edge Cloud. So to summarize, the OpenEbola Edge Clouds are just limited OpenEbola virtualization clusters you might be running already on premises.
07:21
They are deployed on infrastructure of some third party. They are managed fully automatically and I've introduced this kind of buzzword infrastructure service in infrastructure service. Because when you are running virtualization cluster on premises,
07:43
everything is kind of okay for you. You have hosts under full control. You have storage. You have network and choose your addresses. So you don't have any problems. But if you have to deploy the similar virtualization cluster
08:01
on some third party infrastructure, probably hosts are also okay, storage as well. But when it comes to networking, we can expect there will be some limitations introduced by the provider. And regarding the IP addresses, there definitely will be some restrictions.
08:21
So these two things, network things, are kind of challenging parts we have experienced. From the network perspective, it's very environment specific because various providers introduces some various features like a dedicated VLANs for you. But also some limitations like a no multicast support.
08:43
So the solution or the approach we have to take is introduce some kind of common virtual network model which is able to work no matter independently on the provider. So we are using the overlay network, VXLANs,
09:03
but we don't rely on the multicast. So it's just for the unicast only environment. The more complicated thing is the IP addressing. In case of private addressing, maybe we don't care because we have our overlay network
09:20
and we can do anything and nobody cares except us. In case of public IP addresses, the situation is much more complicated because we just can't take our favorite IPV for others and put it into a virtual machine and expect that everything will work. Usually the IP addresses
09:41
are kind of agreed with the provider. You ask the provider for some pool and they give you some addresses back. So this, like a workflow, needs to be automated and in case of edge clouds,
10:03
it's automated through some kind of IP management drivers, which exactly comes to the provider and tells, give me some IP addresses. But that's just one part of the problem. Another part of the problem is that
10:21
you have some IP addresses, but usually you need to notify the provider when you want to use the IP address on the selected host. So that he updates the routing or something to get the traffic to the right place.
10:42
So about the IP addresses, okay, two problems there. The conclusion from this part is simple. If you take some existing infrastructure as a service framework and try to run it within different or maybe the same infrastructure
11:02
as a service, you can't expect the things to work without problems. Now about the provisioning. About the provisioning and how do we, how do we build these clouds? OpenML comes with a set of specialized tools,
11:23
drivers and configurations, which simply talks to the providers and builds all the cloud, just like with a single command run. Mainly we target on the edge clouds, but possibly it doesn't have to be only edge clouds.
11:44
So once again, it's one provision tool which manages the whole lifecycle of this edge cloud. It's command line only. Then we have some kind of integration drivers. It's good to say that when some, like a third party or provider is selected,
12:03
there needs to be two kind of drivers. One driver which is able to allocate some hosts from the provider or release the host back. And the other thing that is mentioned, IP management integration driver, let's say. And of course we have hosts with base operating system.
12:24
We have some addresses, IP addresses. Then the missing part is to configure the hosts so that they can be part of OpenML cluster. So the last part is like a configuration playbooks on the rows for reference architectures.
12:43
That's what user or cloud administrator usually get. But he has to do something. He has to write some provision descriptor which exactly specifies what provider to choose, what are the credentials for the provider,
13:02
what hardware configuration to use for the machines, what to create inside the OpenML or data stores, virtual networks, and so on. And also how to configure the host inside. This is created by the infrastructure administrator and the process is, the high-level process
13:22
is as simple as displayed here. So basically prepare the descriptor, pass it to the tool, wait 10, 15 minutes, and you will get an independent virtualization cluster at the edge.
13:41
Management features of this tool are very simple, very limited. It can create an edge cluster and destroy the edge cluster and the other options are more host-focused, power off, reboot, resume, and so on. So to summarize the current state, we have an integration, I would say a good integration
14:03
with the packet provider and we have some partial integration with the EC2. We have a tool, I will show it a little later hopefully, which can deploy the clouds. It's more like an advanced tool. There are some missing features like cluster can be scaled out or scaled in and the architecture
14:25
which is deployed there is very simple, like a single static one. The future plans, as I mentioned in the beginning, we have received some funding from European Union. The idea is to take what we have and to build from that
14:42
some easy-to-use edge cloud solution which incorporates some catalog for the edge providers and marketplace for the edge application and so on. Mainly that means that we will get
15:02
new integration drivers for new providers. We will get new features like a cluster scaling or a cluster update. The work in progress is a support for light-weight virtual machine monitors like a firecracker and caching data stores and possibly cross-locations networking.
15:26
About the documentation, if you go to the OpenEbla or docs.openebla.org, there is a section which is called this Aggregated Data Centers which describes all these things and tooling and how to write the provision descriptor
15:41
or what configurations you can use and how to parametrize it. Also, I've mentioned the MiniOne tool at the beginning which usually deploys just KVM or LXD single host and evaluation environment, but it can also deploy the edge.
16:02
As seen on the screenshot, it just needs some parameters like select the provider and give some token and project and it does everything automatically. This is not something like an artificial thing,
16:22
but we did some use case validation demos. The most important or the most interesting thing is the video gaming when we have used this tool to deploy around the world 17 edge locations which were like small KVM clusters
16:42
and run on each location one virtual machine which was running inside Wolfenstein enemy territory game server and then from the office we connected to random one in Sydney and we could play and it worked simply. It was as easy as running just a tool and waiting unfortunately 25 minutes
17:03
because some locations like Japan took more time than some other locations which were much near. We have this nice data sheet. We describe all these demos.
17:22
We can give you if you are interested. Also, we have stickers. If you are open about user or just interested, come to us and we will give you all like details you would be interested in. Maybe we have some time for demo.
17:43
Firstly, I will probably show how the provision descriptor to build the cloud looks like. It doesn't have to be necessary to understand all the parts, but just the concept of what is necessary to specify.
18:01
On this page, the most important part is the playbook, exactly the configuration applied on the host. Then there are some defaults which are specifying like a driver and credentials. We are also choosing the bare metal hardware type and centers and so on, but in the next part,
18:25
we just list what host and the number we want to deploy. Then we specify data stores which should appear in the opening block and the last part are the networks. Exactly, this is the thing which creates
18:41
the public IP network with the IPAM packet driver and it requests two IP addresses, two IPv4 addresses from the provider. There are some private networks. Basically, this is enough.
19:01
I've started this provision command before I started the presentation. You can see it took 20 minutes, oh, sorry, 12 minutes to deploy two hosts, to deploy three virtual networks and some data stores.
19:26
Maybe I can try to start some virtual machine. I have to do some, like a workaround to make this working
19:43
because I have Alpine image here locally on my laptop and I just share it to the edge cluster I have right now deployed. I go to the open table. I'll make it a little smaller.
20:05
Yep, I'm finishing. And the thing is that I will run this Alpine Linux on the edge cluster I have right now deployed. It would run as is, what is the interesting part
20:21
is the networking. So I will specify host on the network not to break it and I will give it an alias for the public networking. So maybe I can try two virtual machines
20:51
and if we wait for like 20, 30 seconds, it's already deploying.
21:00
We can check what is the host exactly, okay. We can check but trust me, it's a packet hosted machine in Amsterdam and maybe I can show the packet dashboard.
21:21
It was updated so you can see this is the FOSDEM node, 75, 77 which is exactly what is in here. Unfortunately I think it's copying the image.
21:43
But if that happens, we should be able to ping the virtual machine there. So and that's the goal, that's the goal. On third provider, on third parties infrastructure, be able to build the KVM LXDA virtualization cluster
22:04
we are used to and have most of the features we are used to from the on-premises and also be kind of integrated with the provider, have a public network working and so on. So yeah, let's check the first one.
22:21
Just it's booting and I can just try to log in. Wow, I could log in there. I can check the ifconfig and I can see the private address which is assigned inside is the very same, it's 0.3 address.
22:41
And trust me that it's the Alpine Linux I just deployed. I can see it's 0 minutes. So that's it. The very last thing I will show is that the way I've created this virtualization cluster in just 12 minutes, I can destroy this cluster
23:03
the very same way. I just need to remove the hack I did previously and trying to list the provisions, delete it.
23:23
And it won't work right now because we have running virtual machines there. But I can ask it to clean up and it simply terminates the virtual machines, both. And when it's done, it cancels the hosts
23:41
and releases the IP addresses back to the provider. Let's check there are no hosts and we can check the packet that everything was released. They are not updating it and here is nothing.
24:02
So as said, we have this data sheet here describing the use cases and plans. You can talk to us if you are interested. Also we have stickers, please come to us.
24:22
Thank you.