 Yeah, welcome everybody seems the keynote has ended obviously there are still People coming in That's how it is at this conference welcome and good morning everyone I'm talking a little bit about Linux kernel and what's happening there and how that might be relevant for you It's actually I think I'm doing it for the fifth or six times here on this conference And let's go straight in the quick orientation. Where are we? This is how our colonel that looks like Right now actually the latest kernel is 4.4.5.4.4 team and that's actually a long-term kernel That's a bit different than the other cons kernels you that get released good get released It's get supported for two years actually Most of the recent long-term kernels in the end actually got support for six years It remains to be seen if that will be true for the Linux 5.4.0 also all the other kernels are normally just supported for around about 10 to 13 weeks like 5.5 which is on the under development releases expected on February the third. That's basically nine days from now and Actually, it could be a week earlier. So it could be tomorrow night actually and Or maybe a week later That always depends on how things evolve and what what kind of feeling Linus has how things are 5.6. Thus is expected in mid-April. It's likely a little bit too late for Ubuntu 2004 the next long-term stable release from Ubuntu And and it's might be too late for Frador 32 Both get released a little bit later than mid-April But it might be too too short time to actually integrate it depends on what the developers actually do and I can predict that easily because new releases Nearly all of the time come out every nine or ten week. So it's pretty predictable It's actually I think it's April 8th or April then 13th where when 5.6 will come out and it's possible because it's actually It's a well-tuned engine that's something Jonathan Corbett from LWN net once said Because everything runs straight ahead Each car new car really sets about thirteen thousand five hundred changes, right and brings about three hundred lines of new code actually, it's normally like eight hundred thousand and new lines and Five hundred thousand go out If you look at this this that so you got an idea how much change there's happening A few recent happenings at this point last year I showed a slide that said there were a few hiccups recently And so much for the world tune because it was really a little bit a Lot of things were happening then especially in the media covered a lot of things in the links gone There was a breakage with set of us on Linux and there's a big performance problem. There was a corruption block When when writing data and actually especially the code of conduct and Linux break in there in the fall that was quite quite a lot of In the media But this year I can say there are no similar disturbances in the force recently And code of conduct that was much discussed Ten months ago, it's nothing much heard of so it's everybody said it might be a big big problem In the end there were a few incidents that got reported how the code of conduct says it has to be reported and Those are if you want to See the details about that you can find them on colonel dot octree stays Actually, there were only five reports in 2019 so it's really not that much of a problem as many people fear that might come be There's one thing I need to mention here there was some disturbance And in the kernel world recently, that's more for kernel development. There was at the linux plumbus conference Talked by by a kernel developer that is very active in the security Area He actually works for Google He talked about what's wrong with kernel developer development There are so many things that are not working good like what did he what is this he mentioned your patches get lost and lots of duplicate work and many many other problems and Yeah, if you're interested in details and check out this LWN that article and It actually links to the slides and if you want it really detailed there's actually a bit of it in the video of this session available on YouTube and Thanks to talk. It's actually a work group Was started to actually bring Linux kernel development a bit forward and make it more modern To make it more So it works better So the basic goal is improving and defragmenting the kernel development workflow There were some results already and that there's another LWN that article that goes into details It's a lot of work ahead. And so I won't go into the details here because it would be boring And it remains to be seen what comes out of it. It's likely looking better in the end there's actually some people likely won't like that much Garrett instance now since a few days way what which developers can use to submit pet series and actually Look at the changes even better than than it's possible on the mailing list it's Remains to be seen how current developers will like that and What optimizations need to be done? So this really works out in the end and makes kernel development better And but now if you want to make kernel development if you want to see that on github or github that won't happen I'd say maybe some subsystems will use it a little bit. That was always discussed here and there but Linux likely Won't do that because kernel development is really email driven and Switching to such a far like that would be kind of disturbance. That's not going to happen at least any time soon There was another bit of a recently That was lino saying don't use set of s. It's that simple. I think you've all heard about it set of s is this File system and volume management integrated solution. It was developed for salaris. You can use it on on Linux with set of s on Linux As I said, most of you have likely heard about it. I won't go into the details. There's one thing I wanted to point out He said something about a set of s there a lot of people from the set of s community didn't like He actually said it's not maintained and things like that When he said that he actually meant as a set of s from our girl here Not the our open set of s that's used in some bsd Operating systems or set of s on Linux and he clarified that in a different post. That's linked here so lots of the media coverage didn't mention that and that Really so the articles you read on the net and not really Appropriate for what linus wanted to say But the main message is there's no hope for a set of s merge in linux anytime soon And that's also expected with the well-known Sensing issues that's around them Because the set of s code that's coming from solar actually is on a CDD L and that's likely Incompatible to the GPL used by the linux kernel, but it's good that you clarified that that it won't happen and yeah, and Even if this listen lies licensing stuff gets resolved. I think there are even way other problems ahead That would prevent a quick merge. There were actually some patents There was a lawsuit between net up and Sun and Oracle that actually was Settled in 2010, but those parties didn't say How they settled it or how they agreed so maybe Oracle can't even change the licensing and to make sure it gets and Linux compatible we don't know and there's another problem. There's actually code quality and interaction Because the set of s and the set of s on linux stuff does a little a few quite a few things differently than linux file system Normally do so the kernel maintainers in those areas like file systems and block storage likely would have to what we Say they want to see some changes before that gets merged. So my Opinion there is better stay away from Set of s on linux for now Obviously, you can do that if you want to but if this all this stuff never gets resolved You have to you likely have to switch to butterf s or some other file systems we never have heard of until today sooner or later and If the situation change you obviously Can we evaluate and maybe maybe? Investing but butterf s or some other file system would be the better approach anyway because that's what we have already and Yeah, that's of benefit for linux Already and maybe Fixing all those little things that are not yet get perfect in butterf s butterf s would be less work than bring Set of s into linux is hard to say So that was the end of this stretch a few major happenings that are Right now up front at the wire guard support expected in linux 5.6 Now wire guard is actually a quite a promising and much-praised Tunnel solution for VPNs Linus actually praised it quite a lot in 2018 and There was a It's a state list. It's really easy configuration quick reconnect a lot of a lot of people that experimented it or used it liked it very much and The thing is until now you have to recompile a kernel module with all this wire guard stuff and that's hard and That was the reason by one because there was is Crypto library that the wire guard relies on it actually was developed especially for wire guard and that was called or is called sink and The crypto developers in the linux kernel didn't like that much much, but one of them actually stepped up and finally found a proper solution that was merged and For 5.5. So this gets so the basic problem is out of the way It's actually called franken sink if you wonder because they are it's From frankenstein monster because the zinc and some other solution were merged Zink was rework a little bit and that's what's called franken sink Yeah, but that's in 5.5 and in 5.6 wire guard support will follow So it will soon show up in in the linux kernel and linux distributions Another major happening real-time support soon gets into the mainline when exactly is Remains to be seen and those of you that have been here last year actually might remember. I mentioned that last year already So I was kind of wrong there But yeah, we get to that. It's not I was not totally wrong if you're wondering what the real-time support is therefore it's actually for the industry usage for linux for example to make your laser cutter and Machine with all the machine we used to build cars or something there. You often need real-time support to make sure the machine reacts in time before you have to throw the way the parts Because the laser reacted to last to slowly or something. Yeah, but industry usage is one of the drivers But there are other uses as well And as I said, I mentioned that last year already I was kind of wrong with that because not everything is there yet But kind of right because the option to actually enable this is actually in the linux kernel support already it's called config preempt RT and That happened in 5.3, but there are some things this option depends on so you can't see them actually and Stuff it's not usable yet And that's actually I think only one major thing missing there. That's reworking printk That's the stuff the kernel uses internally to lock something and that's actually in the works There's an LWN that article about it If you want to know more details about that there's also linux plumber conference sessions I think that's available on YouTube and It's not yet really foreseeable when this printk rework will actually get merged looks like 5.7 It seems to be the earliest possible target But maybe it will take two or three more releases more It's hard to say because it was only sent to the kernel mailing list once in December and then since then Nothing much happened. So there's a still bit fine tuning needs to be done before that can can be fixed But the road to actually get this into the kernel Is clear now after some discussion that happened on the plumber's conference last Fall last September to be exact if you want to know more about that. There's a great video there from the kernel recipes in Paris and from Steven rostered and The link is down here. That's just that explains the situation and what you need to do if you are interested in RT development to make sure that your apps work with it and your kernel drivers work for this and Just a small reminder Real time doesn't mean faster in the end means predictable. So for the standard kernels you have on Ubuntu or Fedora, it will likely not get enabled Because the throughput throughputs gets a little bit down If you use this option and so for default kernels, it doesn't make much sense But as an alternate kernel, it's likely To get offered if that's got more like a main light Another major happening that's happening quite slowly is the BP P F or sometimes called e BP F that's changing the Linux kernel slowly, but in really a lot of ways and That's why a lot of things don't notice this one. So what is the BP P F? It's a programmable VM in the Linux kernel Vm not like a virtual machine you use for VMware more like a VM you use comparable with the Java VM, but quite different again and You can do that to get to write programs that are executed within the kernel and That can make things more Easy and quicker because some data doesn't have to to get moved from from the kernel space to user space It's like one of the oldest users who have the old BP F that was a classic BP F It's called these days was TCP dump to make sure you only The kernel only forwards the packages to user land That TCP dump is interested in and that really needs to be done with something like to be BP F As it's way too much data to copy everything out to this program Yeah, but as I said, it's used in many other areas these days right now it's mostly shows itself in the TC that's the traffic control stuff and XDP that's a network fast path that was added like two years ago to yeah, and some tracing and performance monitoring tools BP F compiler collection and BP F trace and as it's a lot of redhead people here actually it's Some of that is supported in redhead enterprise linux now with 8.2 I think the traffic control feature for the net for steering the data was already supported in 8.1 but with 8.2 the which isn't beta right now the BP F compiler collection and And its library will be fully supported in redhead enterprise linux The BP F tracing language which you can use you can It's kind of a detrace like solution to Build small programs that get you the data. You're only interested in that It's still a technology preview and the express data pass is also still a technology preview But I suppose sooner or later that will change if you're interested in this performance stuff, I mentioned with PCC and BP P F Trace And that's actually a great book from Brandon Greg BP F performance tools and it's quite a thick book. I haven't seen it already, but it gets all the details how to actually use it and a few of you might likely have seen this Diagram here on the right side Down down there that it shows where you can actually The lot of small Utilities you can use to look into the kernel and what is done And if the black ones are the older tools and the red ones are new newer more Capable tools that use BP P F or and all that stuff to get more data more efficient without With less overhead to make sure if you're doing performance analyzing or tracing that it's not Impact so much on on your workload There are a lot of small improvements in BP P F land that promise even more usage and more improvements that are happening or Happened recently or happening right now. Here's a small list of those things that happened recently I won't go into the details what what each each of them is doing because that would likely be boring to most of you and Just to quite sum it up The thing is together. They are likely are kind of a building a building blocks for quite different linux column and Also, they will have an impact on the linux os and to illustrate this a little bit more how the impact is if you look at the Defconn's Schedule and search for BP P F you will find What doesn't talk nearly a dozen talks that have BP F in the description already So you you can see it has an impact on a lot of different areas. It's tracing and Container stuff And so there's really happening a lot actually It might turn out to rebel on the ocean eyes linux a little bit and Remains to be seen how things evolve on LWN that which some of you might know is a great website to to check what What's happening the linux kernel? They're actually they actually get mentioned already. There's a microkernel stuff it makes linux more like a microkernel and Actually is someone someone from the red hat that's working on network stuff actually also Recently mentioned. It's let's linux becomes more like a BPF microkernel Thanks to all the recent improvements That are happening there. Yeah, sadly the two things I just mentioned So the real time and the BPF stuff do not work together You fix you fixed what are you fix? You fix it. Oh, I didn't know I know that yeah Okay, yeah, that's this is David Miller. This is your Person you need to thank for for all the networking stuff That's happening with your machine because he maintains the linux kernel Network stack and as he said here he fixed it or I think some fine-tuning is needed Then okay The audience for the records here the audience mentioned that there's also problems of trusted boot and BPF but that's always an issue when there are some new Revolutionizing techniques there are problems and most of the time solutions can be found to work that out that was actually the first part of the talk and Like to give you the chance to ask some questions in the minute a Quick introduction. I didn't do that in the beginning to not bore you that much there Who am I actually? I'm not a kernel developer. I actually I'm a journalist for living I write a lot of things about the linux kernel in a German magazine Actually, some of that stuff of years ago got translated to English, but that sadly stopped I did some regression tracking work a few years ago Sadly, I did that actually in my my spare time sadly this bad time wasn't available anymore. Maybe that Changes soon. I hope so actually and a few people actually might know me from Twitter because I'm twittering lots of Stuff that happened happening in the Linux kernel area I think are things that are close to the Linux kernel and I did a lot of things For fedora, that's also a reason why I'm actually on this conference in particular. We I did that when things like that were Up to date and everybody liked that that was like 15 years ago or something The older one from you might remember that that everybody found that cool and now Nobody has anything like that today and I'm building Linux kernel vanilla packages for fedora's and any of you It's interested in them. So from if you want to check if a bug is happening due to a patch in the That the fedora kernel developers added You can use these kernels or you can use these kernels in from those repositories to actually run the latest mainline kernel And to check out and help develop and developing and checking it and testing it without actually building it So any questions on the topics so far that gives you and me a quick break We have a mic here if somebody wants to walk around As I said, the main things are the code of conduct to remember The workflow ZFS real-time BPF and we have BPF experts here. So if you have any questions There's a question With you, yeah, if it's a call it's a common way for the mic that makes everything easier So it's more like a comment for ZFS first Who is using ZFS in this? in this audience Okay, so now a few One of the things that was not raised on on the mailing list or anywhere else is that ZFS is portable I can collect my ZFS volume to this Mac laptop or to a Windows box and Beta FS is nowhere close to to this So some effort in keeping ZFS just for that. Otherwise, you're stuck with XFAT whenever you want to exchange volumes, which is a pain Okay anymore comments or questions What's the state of shifting file systems for user namespace? Actually, I don't know. I saw that James submitted it in December, I think Or to explain it then ask about shifting file systems. That's important for container usage to make sure Some to make sure that UIDs Don't get mixed up in the containers and the host ID host system And I think James submitted this I saw a few questions about it, but I can't say is anybody in the audience that followed it more closely than I did I don't expect it to get much anytime soon and It's a complicated thing and it will take some time. Sorry Okay, what about the bus There was this pay D bus a few years ago And it was an interesting concept, but the developers abandoned it developers actually Build something new in user land at steep bus. Can't please help me anyone in both name. What was the name D bus? it's What no, no, no, it's it's it's it's a different than D bus implementation that performs better and actually is used in fedora 31 or 30 30 already, I think and Developers that were behind those efforts Drove that but these days they moved on and do something different. So it seems they are satisfied So I don't would not expect any IPC mechanism or more D bus more work on K D bus or something in the Linux a violin violin is more more like a user land stuff that Used is it actually used over D bus? I don't know. Yeah, it's not not something that compute competes in that space Yeah, so let's move on. I already have half an hour used actually Yeah There's slot behind this is free So maybe I can talk a little bit longer, but I don't want to keep you all here But there's yeah to move on the second part will be a bit more about It will be a little bit quicker about recent changes and things that are work in progress that are either Useful or helpful in day-to-day, and that's why I want to tell you about them or good important to know and One of the useful helpful things that is this Is happening in the file system area and As you might know in Linux, they are made basically three major file systems but of s X for and X of s and I'm not talking like then watch and Let you all stand up and repeat something But I want to get make sure you are a little bit awake and want to ask you a question about it Before I do that quick explanation copy and write is something you can use to Copy big files or directory trees in a fraction of a second That's actually one of the cool features from the ZFS I already mentioned and It's supported by the rafling feature and it's interesting for snapshot chain provisioning container images sharing files and like like that So raise your hands if you're listening Okay, I could should have checked how many did that in the first second. So Raise your hands if you have heard about the application or copy on with copy on right. Okay, that makes about 70% I'd say So if you think butterf s supports it Okay, that are just 1015 percent XFS That are less like like the 10% X4 Nobody yeah, it's better than I expected and butterf s and set of s actually support this it's really cool to to copy big big directories and X4 does not and I wanted to see with this small experiment if some of you know about that XFS learn that recently actually And the little scroll side that happened during the four four dot X and five dot X Days with a rafling feature and it happened in small steps and not many people and notice this But it's actually default and productionary since XFS box 5.0. That's actually came out of a few months ago already But it didn't get much attention in the media that this is possible to do now with XFS these days If you want to know more about that there was recently A block post from the Oracle block from from the XFS main developer where you can Find where he explains this how to use it and explains the status of it By the way, if I see somebody Picturing it the slides are on the on the schedule and today you can download it and follow all the links if you're interested in them And I mentioned Oracle there Oracle was kind of a driving force behind that but XFS is also the main file system on redhead enterprise Linux and there it's Supported since rail 8 already. So My copy on right feature is Something that's used already But on the other hand, I'd say it will take likely years Tell a lot of people know about that. It's Bit like the LLVM snapshots are very slow thing those from the device map and LVM crew here Remember that was used to be the case until like five or ten years ago And then some new mechanism showed up to and make make To make to improve the situation, but a lot of people still think LVM snapshots are very slow Even if it's untrue and that's likely the problem of this XFS fade feature as well Another cool feature XFS gets these days is online FS check. It's nearly stable these days There was also a block entry recently that explains the status about this and How to actually use it? Yeah, that's all URL is there also So another helpful useful feature this time not for the enterprise admins here more for the users Let's use Linux on your desktop expert support is in 5.4 and that's makes it easier to mount your SD cards from your modern cameras like video cameras and because they often use XFAT, which is a Microsoft file system But there are some patterns involved. So Nobody wants to touch it in the links kind of but Microsoft has recently stepped up to solve this situation and thing is in Fedora this module is not Enabled right now That's likely because also still due to the pattern situation I guess once Microsoft ships it it should be covered and then Fedora might like might Enable it other developments other Linux distributions are likely not waiting that long There's actually what got added is an interim solution. There's a better implementation That's what's called SDFAT in the staging tree. That's actually really close to getting merged a lot of One crucial developer act most of the patches already, but I've ever found a few problems and Might not happen in 5.6, but maybe in 5.7 this better improved driver gets added so what else there was a Another file system edit in 5.4 That's called Virdio FS and that's a way more efficient way to share files between host and guests It's a bit similar to 9p. Some of you might have heard heard about it's actually used in these cases right now quite a lot But it's actually promising. It promises to deliver better performance There's still up some optimization work that needs to be done Yeah, but it's that's happening the code is actually based on partly based on fuser stuff if you want to know more about this and there's actually a great website That explains things like all the details Sadly the stuff the support site in QMU is still Work in progress unless somebody says it's merged Yeah, it's pretty pretty clear somebody. Yeah, it's pretty close to get merged. I think there were 100 patches So well, but it seems it's not that far away to get much so hopefully we can all use that easily to share files between hosts and and guests assume So what else is there? That's actually I Oh, you rain thing that makes us I mentioned that last year already This is a technique that finally brought proper asset Chronos. I owe AI all for short To Linux last year. It's actually easy to use at least compared to the other AI all things we had in Linux and works a lot better and way quicker and It's actually something the Linux world has kind of waiting for quite a long time because in the Windows world AIO is actually quite standard and that's always a little bit for some developers Find that odd that it's not that the Linux world still uses synchronous IO and Yeah But that's about to change maybe It that used to be bad in Linux because their current support wasn't the best there, but this new solution actually improved things quite a lot and in the past few months got a lot of fine-tuning and a few more features and Also was praised quite a lot And if you wonder why is that relevant because SSDs get getting faster getting crazy faster and basically to use all the power they offer you need to Have a synchronous IO. So if you're doing Application development that really needs IO performance consider using it If you need high throughput or low latencies, that's an interesting solution. There's a talk with all the details and from the current recipes again Against expo, is it actually pronounced like this? okay, okay, and They find all the all the details And actually there was a LWN that article yesterday that explains some of the recent features and and all What what what's possible with this AIO interface these days if you're interested and Look at it. It's still subscriber only for now But in 10 days or something it will be free and the fun thing if you didn't believe the BPPF thing stuff I mentioned earlier on LWN that it's becoming a kind of running gag to Mention things like this where where BPPF is offered a solution For problems developers face in other areas. That's the same here with the AIO stuff And remains to be seen if BPPF will be used But as I said, it's becoming a running gag because it's used in and places where nobody expected it a few years ago So another useful feature that's again more for the administrators here There's a new IO controller. That's called block IO cost that was added in 5.4 was used to be called IO weight So it's way better and more flexible IO controller if you have your multiple containers or VMs and want to make sure none of them can can starve the others by Doing lots of IO then this is important. There were other IO controllers, but One of the things the new IO controller does better is it only limits when there's actually contention So you can't use all the IO you want Until there's some contention and then it actually limits to make sure that your important applications get enough IO One thing to note here is that it's a c-group 2 controller So the more modern version of c-group 2 of the control groups that are used for resource isolation and and think like that And So you really have to use c-groups too, but that's happened the move to see groups to finally is happening after years of production It's actually used in federal 31 already Well eight actually has it and if you want to know more about this there's a Talk about this tomorrow. I don't know in which room brave new world of unified see group hierarchy That gives you a lot of details I expect so Another useful helpful feature that's about to get much Linux kernel 5.6 likely will get us before support. So it's now check again. Who's awake? Who cares? Hands up and not that much thing is For the record it was like five hands or maybe five five percent or something the thing is all of you should care Use before that sounds like a new technique That's maybe come someone and you don't care much But the thing is it's not that far away the next generation of Internet processors for mobile and desktop So the core i11 series or whatever Intel will call it and let's expect it to have use before Because it's basically based on Thunderbird 3 and Intel had already controller So it's not that hard to implement for them and the thing is so you have this hardware coming out This year summer and in a year or two. It's likely a pretty standard and as I mentioned earlier Ubuntu 2004 it's what likely won't use 5.6 yet and that becomes a problem because maybe in a year or two from now You have was before is quite standard You want to boot your USB key to install this Ubuntu and it might not boot so because a proper driver is missing some of you might remember that that happened with USB 3 also and Yeah, that's why it's important that the support for the use before is getting into the colon now because just Like some distribution that I've come come out now are still relevant in two years from now So to be fair Won't will be this point releases film from you Ubuntu They will likely then offer the USB for support. So it won't be that bad but the server still will still use the Colonel that's from the stock release so there could still be a problem So another useful feature that's more for those that are running Linux on on the desktops The support for recently released AMD CPUs GPUs got a little bit better There was a new ready on the series introduced this summer. That's called 5,700 It's actually supported in 5.3 and Mesa. So if you have a recent Linux distribution Like Fedora 31 things will just work even newer cards like the 5,500 work if you install all the updates and And there actually is a 5,600 card that was released just a few days ago. They you need to 5.5 again But it's not that hard to to add that to Fedora if you want to yeah and Also regarding to graphic supports on desktops the driver for these and older cards from from the AMD Radiance series actually is going to support open gr 4.6. It could can be important for some games But they're really thing to take away from this is AMD finally really catch up and the support for for for the Recent AMD graphics is really good these days another thing from the Good to know series is support for Intel's next-gen graphics is also getting merged into the Linux kernel these days into a kernel and Mesa and Intel's graphics drivers also Support already support open GL 4.3 4.6 after preparing working towards it for like two years, I'd say and So that's a major achievement layer Yeah, and I mentioned two drivers here actually the 965 is an older open GL driver in Mesa to support the recent cards There's a new one with which promises to be to offer more performance and to be lighter That's not the default yet But it looks like it will take over soon with the next Mesa release that Contains all the 3d drivers you normally install on your on your laptops together with your distributions So in the end AMD and Intel have quite good Floss driver stacks these days. They are not perfect But maybe it's Simply like that. They are so good. Maybe consider avoiding NVIDIA if you can While talking about NVIDIA a quick note, Nouveau the free and open source driver for GeForce cards still leaves a lot to wish for That's like it for years now for example with the latest generation the two-in cards that for example, there's 1660 and the 2000 cards from the GeForce line You don't get 3d support that might make even your GNOME desktop and other desktop environments slow And that's because there's not really Distributable firmware from NVIDIA yet. It looks like this is about to Change there were some patches recently that are also expected in 5.6 That fixes, but I haven't seen the firmware yet so remains to be seen what's coming out of it and The biggest problem for the Nouveau driver still remains where it is the driver simply cannot reach Cannot switch the GPU into their highest speed states or the lowest power states because the Distributual firmware that NVIDIA offers to links to Distributors to integrate simply doesn't support that so this free and open source driver Will never get the speed you get with the proprietary driver for now. So that's really a pity maybe it's about to change there's NVIDIA developer conference Technology conference and in March, I think and in a schedule There's actually one thing that Indicates a little bit that NVIDIA might change there Their strategy and get a little bit more open I'm not so sure if that will really happen We have in the past few years every there were every now and then some indicators that Intel That NVIDIA would become a little bit more open in the end They did quite a few things to improve the situations, but on the other hand in some areas it got worse again. So it In the end got really the situation wasn't a lot and so I'm a bit skeptical what comes out of it Yeah, let's wait for March and see or maybe a bit more and yeah until then for now We have their good AMD and and Intel support So one more good thing to know you can use C-Lang Compiler and a C compiler that's built on LLVM to build kernels now on x86 64 so 64 Bit x86 86 machines most of us have at home. It might still be a bit bumpy But it can be of interest to use the security features that are in in C-Lang and LLVM If you're interested in that there's a website about this that has actually an issue tracker Where you can check if you run into issues if that's something that's already known And if it's something that needs to be fixed on the car or side or the C-Lang side Another thing good to know is PSI monitor That enhances the pressure stall information PSI stuff that got added in 4.20. I talked about this last year When I was here the PSI stuff is basically product load average on on steroids Because you can see a CPU and memory and I owe in your own lines to check if how the Pressure is and how busy they are and I won't go into details if you're interested in this Simply look at LWM net or Google for it last year's major Hayden After my talk wrote a blog entry about it that has it's easy to find the details there But this PSI stuff got enhanced that as I mentioned it's a PSI monitor stuff that allows user-land stuff to use them programs to better monitor what's happening there and that actually allows way better out of memory handling and can make you help your Desktop shell from prevent from star stalling because in the background I always happening and They're actually four solutions built on this One one is actually an Android one from endless The Facebook developers that actually that develop here PSI actually have something built on that as well and the gnome camp has this early OOM That promises to to make sure the system runs Better and the letter was actually proposed for for fedora's party to for integration that Triggered a big discussion on the mailing list so Because some people would prefer this the solution from Facebook. That's actually as that might be added to system D directly it's so open so Looks right now. That's for dollar 32 won't get anything any of this and but Yeah Maybe 34 is 33 or something then we'll see Another thing good to know be cash FS. That's also a file system. That's kind of kind of competing with butter of s and set of s There was not much progress with their recently It will likely take years till till ready if ever It's still a one-man show. So I I'm a bit careful about that how that works out and Be aware it has a loud fan crowd already really a loud fan crowd. Maybe it won't be as good as anybody expected to be and Hard to tell right now because as I said, there's not much happening. They're just one developer Remains to be seen. He wanted to get it merged into the kernel that as he said that two one year and Three or four months ago, but then not much happened since then Maybe but of s is better suited by now is sound a little bit like a but of s fanboy, which I'm not it's just how it is Hard to say so and he said I'm getting close to the end And I think good to know is the limits going to get multi-pass TCP That will be in 5.6 and to that was basically done here on the conference because Dave here Did this yesterday? I think he merged it to the Linux and that next three that will wear stuff for 5.6 Is prepared and he only did it for the Twitter double-mind. I Think I looked earlier today. You had a 250 Lot of lot of likes and yeah, and that's Yeah, the Twitter handle is mentioned here and give him give him some hearts Yeah, and that's actually Important if you're having going out with your your laptop and out of the reach for the Wi-Fi and then it can switch over to the mobile connection and In the background and your apps won't notice any of this So it's really practical especially for your for your smartphones And that's actually the reason why Android has something similar already and yeah So another useful feature. I think it's the last one on my slides support Raspberry Pi for Got gets improved right now as you might know the Raspberry Pi Is not using the after him kind of use a special kind of it has lots of added drivers But some developers make sure the mainstream kernel the mainline kernel gets support and 5.6 It will improve quite a lot and if you're wondering why is that important to me? Yeah, it's important to make sure You know Fedora your Ubuntu runs on your single board computer Out of the box better There are lots of other stuff Things I could talk about But I guess a lot of people I have something for the next time slot. That's why I'm not going into the details here as I as I said this time slot right behind this is Is free maybe we can get back to this in a few minutes But first I want to get a few questions ask if you have them and do the summary the summary is Linux kernel is a fine-tuned engine lots of things happening fast development It's really great the code of conduct that created lots of trouble 2018 is not nothing much heard of this here Viagard is approaching real-time is approaching XFS now supports the application for fast IO consider using this new AIO technique the IO u-ring stuff and Those were just the highlights. I mentioned and obviously there's better hardware support also And the really big thing is the but BPPF really changes the Linux kernel Ruins to be seen what comes out of it if you want to know more about this follow LWN nut That's a great site that covers all the things I mentioned here even in more detail You can consider following me as well on on on Twitter I have a car a count that's called Colonel logger Well, we always talk about things like that. I actually have three as six or five other accounts one of them actually is dedicated to to the redhead and Linux redhead and federal stuff I do and to keep things a bit separated and yeah And one more thing Provide feedback talk to me if you How this talk was what I should be improve could improve last year There was something on on the schedule to to report feedback. I think it's not there this year So, but I guess the organizers will ask for feedback anyway That's it. It's actually slide to 183 if anybody wonders Yeah, but are there questions Hello, I'd like to ask about that out of memory handling if I have a program that keeps lots of data in cash and The om hander would kill the program. Is there some way how it could tell the program to Release the cash instead of doing the program To release the cash if I have a program that has lots of data cash I think from the Android camp there was some patches to Improve the situation where the current can talk can ask user lands Yes, use a land apps to release memory. They don't Yes, and I have this Tom stoning mechanism Is that under any of these tops Tom stone mechanism where it basically are the main cause to Save the information somewhere Let's talk about it afterwards And I think the solutions I mentioned as those are just user land solutions So that's you can actually program that and So can this user and solutions ask the program to quit in an auto release memory instead of doing it Do you mean if the programs can do that? Out of memories user lands demons or the colonel the demons of the kernel I don't know which one, but it will be nice to get some feedback Instead of being killed the short answer is Colonel develop us actually from Google are working on this and I don't know what the status is But I think it's getting possible. Yeah Any more questions Right one at the back What we're going to carry on with questions beyond 11 o'clock So if you need to leave for another talk do so but the rest of us will carry on So I was a bit concerned with your comments regarding USB 4 not booting So does that mean there is no competitive Yes, so my question is your comment you commented that USB 4 You know LTS you would do 20 LTS would not boot on you on USB 4 and You mentioned that your you your USB key would not boot At two years down the road because the Colonel did not support you as before Does that imply there is no compatibility with USB 3 or 2? How does this work that work? No backwards compatibility to use if they if there's no backwards compatibility to USB 3 and 2 Yeah, because use be for is basically Thunderbolt and it has it has some backwards compatibility to proper us be But If you have a USB 4 key or storage device That will only work if you have a USB 4 driver and even if you have a backwards compatibility Device that's actually connected to a controller that is almost before only then you need support for that controller to make sure To boot from it But you could at least have the the early stages of the boot which are handled by the by EFI or whatever that part would work, right? You mean the boot stuff handled by the firmware? Well, yes Yeah, as soon as the Colonel runs the colon has to do everything on your own kind of but so couldn't you just rebuild? You know just reveal your any tardy or something like that You basically need to use be for stuff support if it's of use be for controller relative to TCP multi Path this will just left. Yeah, I just want to bring up the point. This seems likely to be you know a Big contentious point because it will drive a lot of you know, sorry This will probably be a contentious point with regards to you know a Multi-wan WAN Router products in the market, right? If I understand it, right? If I have a Linux shop, essentially I can dodge those and Just use TCP multi-par thing to access my multiple WAN links Main Was so loud. I'm not sure I got it So if it's more your question was if it's more for the routers or routers or if it's more for your client My point was when I have TCP multi path. I can dodge my WAN routers So my multi WAN routers, right and can access my white area network links directly Utilizing that new technology or am I mistaken? No, that's possible. That's actually Depend there will be use that's the idea to have a user land Demand that actually controls if you're using both links at the same time One will prefer because you can't do that. That's policy you have to do in user land That's depending. Yeah, that's that's yeah, that's an additional good information, you know bonding to basically max performance versus non bonding and just do the fail-overs the main usage is you Take your laptop and go to the garden where the Wi-Fi ends and then it can switch to Mobile connection on its own. Yep And I would have a question is there a chance that kernel will provide because last year was the year of in those mitigation box basically which do have some impact on performance in some cases and Is there a chance that I would have an access to list where some Performance impacts are like marked that if I disable this mitigation I will get this speed back or Something in that category. So because there's too many mitigations today and one does not know what really is impacting his His scornella. Yeah, that's that there is no I say that's actually not possible because the How big the impact of these mitigations is totally depends on on your on your workload So if you have something that does contact switches a lot then this PTI and this spectral stuff Might really hurt you but if you have not so much contact switches There's no not not so much overhead You don't gain that much if you did Disable the mitigation and maybe the thing is if you're really in a situation where you consider Disabling those mitigations for the CPU box Does it actually matter which of those you? That's what I don't know From the security impact is likely not that much of a difference And in the end you actually have to try Which of those makes a difference and the best starting point for it is actually a current parameter called mitigations Off that this disables all the mitigations for those security backs and Then you can able and able and disable them One by one if you really want to Liberating on that question and answer group of pairs here in particular in the specter Area of CPU box Intel box. There is a zoo of Specter incarnations. So it's Yeah, yeah, exactly. So my point is it's actually pretty hard to make that kind of decision, you know Can I disable like this very specific incarnation to optimize my performance in this? You know context switching heavy situation. It's much rather, you know, do I want to play it safe? Do I want to max out my performance and then I have to basically Secure my system respectively by other means most of these bugs except for the Meltdown stuff are really hard to use and so maybe If you are in a controlled situation where where it's even possible to disable then Maybe Disabling all of them except the PTI Stuff that's protecting against meltdown might be possible But yeah, it's security and if you disable security features You always have to look into into them closely to know what you're doing. Nobody can nobody else can do that for you And that's what I look that the page would like quantify impact and and performance and somehow so because I cannot track all those Tense of mitigation. So like somebody simply needs to sit down and explain that on some Web page, I don't think that's anything that kind of any more questions and if you're interested in any of this or what I'm having What did I do now? No? Just ask Like yeah, no, I hit the wrong button How did I do that? No, you can ask I still have time the next Speakers only expected in 10 15 minutes from now or other questions for the stuff I talked about now Okay, then I leave you to it. Thanks for listening. Yeah, I have a great conference