Jakarta EE Tech Talks, 27 Apr 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Elastic JVM for Elastic Java EE Applications Running in Containers | Jakarta Tech Talks - Oct 2019

Description

Containers provide much better elasticity and density than VMs, but JVM-based applications are not fully container-ready. The first issue is that HotSpot JVM doesn’t release unused committed heap memory automatically. Second, it is not possible to increase the size of the JVM heap at runtime. To solve these major issues and make JVM more container-friendly, a new patch is implemented for the Garbage-First collector in OpenJDK. This session shares details of what is done and how the added improvements enhance Java resource consumption efficiency.

///GET SOCIAL!

Subscribe: https://www.youtube.com/user/EclipseFdn
Follow us on Twitter: https://twitter.com/JakartaEE
Like us on Facebook: https://www.facebook.com/JakartaEE/
Join us on LinkedIn: https://www.linkedin.com/showcase/jakartaee
Tag us: #JakartaEE
Want more great Eclipse Community webinars? Become a member of our Jakarta Tech Talks Meetup: https://www.meetup.com/jakartatechtalks

A

Hello, hi everyone so welcome to the session with Ruslan and Brunel I have to apologize to everyone because, as you know, there is a time difference, change in Europe and my assumption was that were gearing towards North American. The top.

B

B

Yeah, so the problem was that it was time zone change in Europe.

A

B

Yeah, it's a good point to you know to deliver second presentation for North America.

A

I don't take takeaway from from from the time so yeah Ruslan. If you want to take it away, please go ahead and share your screen Bruno. Let us know whenever you want to take over or please let everyone know. When is the good time to ask questions and so on? Okay,.

C

Cool, so everyone Ruslan is gonna, start the presentation a minute here and why he does that. Let me encourage you, that's watching us here to ask as many questions as you want. I'm gonna be monitoring the shots and we want here to do a presentation. That's that answer. Your questions so feel free to go ahead. There's a chat on the side here or on the side. Here depends on on what you're using so you know, take a look go there. Actually, you could start by saying hello and where are you from right?

C

I saw you know. um I saw a couple here that I know here from the US from Brazil. So how about you say where are you from and then we can get started right. So you saying the shots, so hi I see from Chennai India thanks a lot man for being here. That's great! So you know so you tell your so so now you tell you are you from? We know who you are and also you know where the Shetty is.

C

That's all great right: Sankar for Chennai, that's great Yanis from Qatar, that's good, marching from Czech Republic, I'd love to Czech, Republic, I, don't know! Qatar I haven't beaches Chennai, but I love the Czech Republic Marcia from Brazil Rodrigo from Rio. That's very good, excellent and they're from Budapest excellence from Belgium with Carlos from Belgian. So we have people from all over the world. It's great! It's great to have you all here. One awesome.

B

Yeah Java community is all around the world, it's very good to have people around the world and thank you for joining this presentation. So Bruno, if you don't mind like I, can start talking about the topic. Yeah.

C

B

And today we are going to discuss very interesting and the hot topic related to elasticity of JVM and Java a EE application. Now like Jakarta EE applications are running inside containers, so I will be fully transparent with you. This topic is new for me as well. Even I am about seven years in this area. However, its updates are going on. So almost let's say every month we get something new, that I learn as well so and the goal for this presentation to share things.

B

I already learned and collect additional questions, so I can bring two people who developed for the improvements we can discuss the business and maybe improve with even better good. So and of course it's us, you build us. Probably you understand it's quite boring to talk to the screen like 45 minutes. So if you have questions, please feel free to ask them during my speech. It will be much more interesting for both sides so and today, at the in session, we will discuss the following points. What is the problem?

B

What kind of problem we saw in general right and what kind of improvements were introduced recently in Java and how this improvements work together with Jakarta E, and we will go through some garbage collection, testing and actually I'm. Sorry I forgot to introduce myself.

B

My name is Ruslan I'm, co-founder and CEO of jelastic platform as a service for ISVs developers and service providers, so we're working with Java Melissa T a long time and specifically run in containers more than seven years, even like containers became popular like last four or five years, but we had experience before so great. Let's jump to the problems in the survey which was collected in 2018, one of the major problem which was highlighted by developers around the world is large memory requirements and it's true.

B

Everyone knows that Java is greedy in terms of memory usage. It ran at start point and during the run time.

B

So the question is like: can we solve this problem and if we can like what are the possible solutions? Of course, there is a specific reason why Java consumes so much memory, because performance is important and in the past, performance was always number one for many applications. But, however, is any of the options you know like true, maybe to give up for some performance and but get better elasticity. So today what we have actually what we had in the past.

B

You can see the screenshot from this is a visual VM, which is a tool that can analyze memory usage of GBM and, as you can see, three major points to consider like blue line. It's a real used heap, so this is objects that are store it inside the heap and then orange space is how much memory were committed by gvf. So how much memory Julian took from the OS and then like max heaps all the red line, its limit related to memory heap.

B

We specify it through X, amix parameter at the beginning so and what we hear today, what would be actually had in the past that very often orange space get very high, and but real usage is less than the max heap. Let's say right, but GBM keeps a new stream and it does not release back to aggression system in the past when we run JVM, stubborn metal or inside, like virtual machines, and when we run like multiple applications, this inside Banjul IAM.

B

The problem was not so big and people didn't pay attention, but today we are living in another world, so we run many gbm's, specifically a micro service role right like and we can run like thousands of JVM inside one big, but at the core machine. But if all the sounds and gbm's are inefficient, so we lose a lot of money and infrastructure becomes underutilized. So it's not really good.

B

Let's discuss the reason why we seek for JVM elasticity. I will mention specific cases, I believe people will come up with additional cases, but this case is obvious, at least for some group of people who are working on the solution. So number one would say like cron jobs for data processing, very often f-from, chrome Jobs's. It can become active a specific period of time and they can pull a lot of data inside memory.

B

Do some kind of aggregation or segregation and then like store new data in the database, and then they go to sleep mode again, very often after such kind of processing, as GBM doesn't release ram back to operational system, so we have committed, but a new stream inside the Juvia machines. So this is one case. I hope it's clear. If not, please ask question second case.

B

Well, for example, you run like 100 or, let's say, thousand of gbm's inside one big parameter machine or like large virtual machine and for example, at some point, one virtual machine needs more sorry. Gbm needs more resources, so you can give this resources from the common pool for like temporary burst, for temporary data processing or for temporarily lower spike, and then, when lost by goals, you can release this resources better into the pool, so another GBM can get it when it's needed and another example.

B

Some service providers introduced it a previous model. It means you pay a lot for the limits of your virtual machine or your body metal. You pay for exactly amount resources, your applications concealed. So if your application is efficient, I mean like it can release a new stream back to operation system and it's not needed. You can pay less at Z, so it saves money for you and you can reinvest this money in the product, development or just increase holidays military team, so by decreasing money spending on the infrastructure.

B

So, but with elasticity it's possible to adjust a resource usage on the fly without elasticity is impossible, so you have to do or restart on GBM which lead leads to down times and very often down times are not acceptable. No or it leads just to out of memory error or to out of memory until exception, so your GBM will just die. This is a reason why we are trying to bring plasticity.

B

Juvia java wrote: let's talk a little bit about some problems when we face so specifically, what is our memory error and what is om killer?

B

So there are many kind of out of memory errors they happen in different spaces and a good explanation will be this picture, so we have like different layers like operation system with, like shared libraries allocated. Then we have like heap with young, can all generation, and then we have of heap space where, like direct byte buffer and the code, cache are located.

B

So if your today we are talking specifically about heap, of course problem exists also inside of heap, but it's not for this session because again like we did not get to that point where we are working now, only on improving elasticity of the heap so and when out of memory exception happens when blue line cross the red line.

B

So when the realest Resources gets very close to each some X parameter to the limits of Sharia and another possible exception, where native of keep let's say, Ram gets close to the containers limit, so in this case, out of om killer don't happen. So operation system will kill your GBM process and you will lose all your state in this case out of memory.

B

Exceptionally, okay, you could happen so your GBM will stay, but it will be an error inside your application, so you will need to handle this somehow so looks likes to many points to consider right. It's true. There are too many points to consider, however, we believe, step by step. We can clarify all these points and many people will understand better how elasticity works with trivia.

B

In addition to all the mentioned points, I would like to add it's very important. Where is your GBM Runyan, because there are different kind of containers now on the market and different kind of virtual environments? So you can run inside dog here or you can run inside, let's say Alexey container or you show other containers or inside virtual machine butter metal so because Sam container technologies do not provide enough insulation would say so the expose limits of the host machine and it can be a problem for you, so you need to consider runtime virtual environment.

B

So now, let's talk little bit about the paper.

B

Usb link model just again like to clarify why it's important, for example, today, very often intersection service providers sell visual machines and you go buy like let's say, 16 gigabyte of RAM and they always pay for sistemas, while your application can use much less so you kind of building model, your generation will say, building model which is called like pay-per-use offers more efficient solution, so you can set the limits, but you can pay only for what you use and if we, if we can make Java and lastic, so we can say it again like a lot of money good.

B

So we decided to store to start improving the vertical scaling of Java from hip, but there is also ongoing work related to X MX parameter when you will be able to adjust this parameter on the client, for example, when it can be helpful. For example, you did a wrong assumption at the beginning right, it said like okay, my GBM needs like 32 gigabyte of RAM, but in reality, like your GBM, needs more like, let's say, like 4, more gigabyte of RAM.

B

So today you have to restart the whole JVM and again, like you, don't lose the state and JVM will again like read all the states from the disk. It will increase. I hope it will did. I performance will differ update, so it will slow down everything if I bring more elastic solution to enable people just to adjust exam mix of the fly.

B

If you help to solve this problem and the reason I'm going to work on this, however, as I mentioned today, we will focus only on elasticity of heap how to bring this orange committed ram down how to release unused memory.

B

So in the past we did some work around because it was possible to do with specific approach when you explicitly call full garbage election cycle, it's not very good approach because it significantly influences performance, so your application can degradate performance. Application can degrade eight. However, it was only one possible solution so and it looked like this, when orange line grows up to max-heap blue line grows, let's say like till the middle right and then Lord Goes Down and you call like full garbage collection cycle and orange line goes down as well.

B

It was possible to do with specific garbage collectors, not with everyone. For example, Ji Won was suitable for this kind of approach, so we did a lot of tests in the test and we tested all the real applications. We we present the solution, tool and customers and we run production from top of it and we figure out. Ok, it can be a workable solution, so many people are satisfied with this and we decided ok, maybe we need to introduce some punch inside JVM and that's what we did by the way we introduced it.

B

Three parameters as it can configure elasticity of JVM, specifically if JVM is running with g1 garbage collector. So you can follow this link that this improvement or was added to Java 12. So it's available now you can enjoy this improvement and in general it works like this. So now it looks very elastic. So as soon as used hip is not I mean as soon as memory is not needed, anymore, GBM memory usage goes down automatically.

B

You don't need to call like foo garbage collect collection cycle, so you can easily configure it now and again like this is the parameters that you need to take into account, while configuring elasticity for G, 1 and community understood, let's say the importance of this improvement and we got recognition from the community. I specifically want to highlight contribution of Rodrigo Bruno, who created the patch. Let's say who created, who created the specific code for this patch and contributed it to the open source, open, JDK good.

B

Let's go little bit now to real results, real tests, so there are many different kind of garbage. Collectors will go run a set of tests with different garbage collectors to see how they behave and we will run a simple test. So we will just create array with string array and add it to the linked list and do it in cycle.

B

So, first of all, we will test g1 garbage collector by the way, any questions. So far, Bruno Duque me guys.

C

Yes, right now we hear you there's no question so far. Job.

B

C

B

Good so anything happen, just let me know so: g1 garbage collector how it works. You need to specify, for example, + use, G 1 GC, and you specify a periodical in interval when you need to call garbage collection cycle. So at this example, every second I specify like please call garbage collection collection cycle every second so and Jibreel will take at will analyze and if it's possible to do release without food garbage collection cycle it will do. If not, then it will call for garbage collection cycle.

B

So, as you can see, it's elastic now right in the past. It was not like this, but now it's elastic and it's cooler automated.

B

There is additional threshold. For example, if your system is busy- and you don't want to call garbage collection cycle because, like you, don't want to do like to face like throttle degradation, you can specify like specific thresholds, which is related to log average and say like okay. If load Lord average is more than let's say like one or like 0.5, please don't do any garbage collections, so please don't do elasticity, but if it's less than this result, then GM will will do the job.

B

It's very useful again like for productions even right like or for highly loaded stage environments. So you can regulate performance and elasticity.

B

There is an issue. There is one important issue to take into account when you do it inside docker containers, because the steel to date, but our containers they expose limits of the cost machine inside so GM, does not recognize like what average inside container. But it's quite easy to solve. You need to do mount of MXC fest and there is a specific example how to do this. You can go and read articles how it works. But if you do it, then the problem will disappear.

B

So then you can check what average right inside the container so just important point to mention here another one good garbage collector they became it becomes more widely adopted, which is called solando.

B

This garbage collector I like this garbage collector, but it's very elastic. It has different kind of functions, it can run large heap, it can be elastic. So of course it's still new compared to other garbage collectors, but as they know, people committing a lot of efforts to make it like fully fully production, ready and I like it very much because it has specific option, heuristic option like compact and when it's specified, then your GBM will be very much elastic.

B

So you can use it for again like for, for your means to take a look, how it works and maybe to you know like to promote it to productions if it does not influence your performance so compared to g1 I would say it more efficient in terms of CPU usage and even like maximum usage during the spike loss.

B

Of course again like it's a synthetic test, it cannot reflect all cases, so I just must do in order to do to emphasizes because, like there are some cases when I g1 outperform chalindo so but at my case at my specific test, it's different so another one garbage collector, which is quite new and was developed by Oracle for large heaps zgc and recently in G decade, serene Oracle team introduced it same solution like uncommit and used ram back to operation system. So, as you can see, it's it's going on.

B

People are improving garbage collectors and taking elasticity into consideration. So there is an example how it works on the same test. Right there are options you need to specify like unlock, experimental vm options, then use zgc, then uncommit delay and collection interval like so.

B

You can go after the presentation to website of this garbage collector and check specific detail of these options, but they enable in short again like they enable elasticity and, as you can see, our GBM can get memory and really release it back dynamically.

B

Next one is C form garbage collector from other systems. It's a commercial solution and I believe again, like zine, can see, for one was, was one of the first solutions provided elasticity for Java on the market, and recently we got access to the solution and tested how it works. It's very interesting one you just specify like use is a steep, but you need to setup it properly. So there is a specific approach. How you must set up.

B

So there is a specific approach. How you must set up is in GBM I, believe if you reach all systems, they will help you, but in general we did testing and it's elastic. So, as you can see, elasticity is quite is quite good you need to, but to measure it inside containers you need to enable C groups inside configuration files, so I will not go to so deep details. If you have additional specific questions about, one of the garbage collectors feel free to follow up me after this conversation, so another one concurrent mark-sweep collector.

B

Actually it's going to be deprecated, but still just good to understand how it works. It can be elastic as well, but you must force full the garbage collect collection cycle so explicitly inside your code or outside through GK, MD I would say, and even you need to do it like four times to bring it down to zero.

B

Okay. Next one is set on garbage collector same similar to concurrent mark-sweep, so you must explicitly call full garbage collectors collection cycle good. Next one is open. G 9, which was initially developed by IBM and now, is under eclipse foundation this JVM introduced elasticity. Earlier then, let's say some other garbage collectors.

B

However, I am personally facing some issues to configure it properly to measure real usage, so in my case, gim just consumes ram like and it keeps it, and I cannot measure like how much resources was used in reality, I reached developers and they provided some specific script, how to measure real usage. So yes, after I did it I figure out.

B

Ok like there is some elasticity and memory goes up and down bad steel, like my matrix inside container, they show me that memory is flat and- and this is not something convenient again like I- don't want to blame OpenGL developer it most likely it's related to my configuration, so I didn't configure it properly. It's moment, I'm talking to developers and as soon as I get qualification of this point. I will share the details with the community, and just you know just for comparison. Epsilon epsilon is not elastic at all like.

B

So, if you want to do elasticity, epsilon is not for you.

B

Parallel a garbage collector, it's not elastic at all as well like, so it will get the RAM from operation system and if you keep it forever, so it will not release it back so guys as a quick summary, as you can see, there are different garbage collectors and they can work in different way and, moreover, you need to take into account some specific major points of the elasticity. So I created simple illustration or for this like so a what is the initial memory usage when GM starts B? What is the maximum usage for your lot?

B

What is the maximum usage at the largest? Let's say: lot spike C. What is the growth speed so how fast gvm can get from A to B because, like what I saw, some garbage collectors can be faster compared to others so, and it directly includes the performance of your application. So you need to take into account D like how long GBM keeps a new Streb.

B

So let's say like after the Lord goes so because Sam again like depending on the options and the parameters, it can stay a longer or shorter and it depends on your load again like if you have dynamic load and it goes through short period of time. Maybe you don't need to do this elasticity, but if you have enough time, you know like to bring it down I. Can you know like couple hours without load? It can save your some resources, so you can utilize by other infrastructure.

B

So then a b, c d, ee, sorry e release, speed like how fast it release so does it go really fast or like? You need to do like several steps to release full and use memory and f.

B

What is the minimum usage after the release because, like very often it happens that after release memory is still higher compared to initial sites, so take into account this all points play around or test it and provide feedback and maybe even like all together we can make it more convenient and more automated, so I did the same text in Kim, Koo, brilliant! It's during the lack of time I will not go yes, yeah.

C

Before we go drink the kubernetes there is some there's a question here from Sergey and he asks about there's any negative performers effect. So can you talk a little bit about the difference, garbage collections and their performance effects? Yeah.

B

Definitely so very good, very good question guys like if your main aim is performance, just don't illicit. You just get everything Oh Bram inside like and keep it because, like again like allocating memory, is very heavy task, so its CPU, it consumes CPU and it takes time to to get it. So if you, for example, applications is really critical like to performance, just don't do it like so just specify like IXY mix equal like 2x MS I get a ram from the beginning and run it like.

B

This, then, like second option like okay, like if performance is important, but it's not critical for you right. So, for example, because, like let's say like you, can wait like hundred milliseconds like right, like or 200 milliseconds yeah, why, like additional 200, milliseconds right so for like web applications in many cases, is not so much critical and very often again like web applications. They have dynamic load because people do not come to the store like in the morning kinda day when they sleep application.

B

Does nothing so for such kind of applications again like if you properly configure vertical scaling, it should not influence your performance because, like you can again like measure? Ok, it is a bit lower inside inside JVM. Don't do LSD CTO case. There is no law inside JVM like bringing it down, so it's possible to do to find an optimal combination for web applications. Let's say and.

C

B

C

This this this object, optimal, compare combination. Is this basic on, because that way, your application works, or you know like web application, has a similar, similar usage. How do how do you do this Sergey asked if there's a best practice for those parameters and Gordon goes further. If there is a tool or test, they can measure which configuration works best for you is. There is anything like this yeah.

B

B

Very good, so optimal configuration, it depends on several layers. To be, to be honest, it depends on several layers, just a for like how what kind of garbage collector reuse one then like what kind. How do you configure this garbage collector like because different garbage collector provide different options right like, and you can fine-tune them specific for specific work load?

B

Then it goes like okay like how your application works and what is it Lord for now, I will suggest you to take a look, let's say like weekly Lord like or like daily Lord work right like so, you take a look at your daily Lord if you grow up. Okay like this is like maxing all right, and this is a minimum, and this is the speed of the growth like, so people are coming to the website during one hour. That's it quite like ordering like two hours like so the world.

B

The world grows like during two hours like so or like Claude goes like during like 10 second, so it very much depends like because, like 10 second and two hours, it's a huge difference, yep. So take a look at your current Lord, analyze it and then you'll be able to get a better understanding like what exactly you need like how to to.

B

There is no like serial bullet for everything, to be honest, like I, hope and I. Believe and again. This is different reason why we are doing this in the future. We will introduce something like adaptive performance optimization so when an application runs and checks like okay, like I, have this load like and it's better to tune these parameters for myself or you know, and when, like plots by cons like and law changes, the JVM will be able to adjust parameters inside, like you know, like keep rambling gear or like release it faster.

B

So but it's it's for the future like for now. We need to solve, like this simple, vertical scaling way because really like we were struggling with this issue like for some for many years, like all other languages, I elastic like but Java is not like it's it's not fair, it's not sexy, so we want to make it elastic.

B

Good anything else to it to this question.

C

No I think that's fine, there's some other question I'm waiting. You have to finish this for any other questions. Ok,.

B

No problem so guys like just again just to mention I, did the same in kubernetes and there is a link to Google in ATC amo, so you can go. I can see what is inside like how it works. Maybe you can run inside your know, kubernetes and yeah. There is primitives and graph on already I mean like llamo for parameter for parameter was amber fauna. So you be able to see graphics like this. Let's say right, so this is G, 1 and kubernetes.

B

This is Shannon talking good relations, zgc and kubernetes, right like so, and and I have like all of them all together. So they can look at this. It's very useful.

B

Do you know to compare like how all of them are working like and, as you can see, there is a difference like just a phone based on the maximum like Ram consumed by a JVM then like how fast JVM does the same job like, for example, at this specific case like a game like zgc, did longer compared to others, but like because I contributed this in a specific way like I reached developers, they suggested me to improve some parameters like, and it will be the same.

B

So the next presentation I will show like okay like this, is, if I run like the GGC, with different parameters like how it worked with how it influences elasticity so just to again like to mention that there is a specific package for kubernetes and you can play this package then coming to finally,.

C

Since you're just gonna skim, it true is this presentation available, so people people can receive a link about about that yeah.

B

And I will share the link after yeah. We will share the link in Twitter, so please follow our Twitter and Jakarta will share the link as well, okay, good.

B

So next one we need to understand like where the memory can be consumed right like so now we solved it sounds like partially solved like masticity inside GBM right, but like take a look of cost like inside applications. You are very like Zen, like libraries, you bring with your application and inside your called your core should not have memory leaks and it should release unused objects.

B

Otherwise, like obesity will not work so I, please could see the take it takes this into account like so now. Gbm is elastic. The next is your responsibility, so application server, libraries and application so and if you make all of them elastic at the end memory such will be elastic. So we did simple load testing with a arum which is great solution and it's Jakarta EE certified.

B

So the test is quite simple: we created GSP Paige, we just put object inside the session, so we put like string, object inside session and then sorry and then we call the link like this page in loop so and how it works. Look. This is a number you see it's elastic, so now it's elastic in the past. It could keep the bram like this maximum brain like for a long long time. So you don't use you, don't need it, but it's still there. So now it's elastic and second one g, one with dragon.

B

Well actually I forgot to mention another kind of garbage collector, which was developed by Alibaba and Alibaba recently introduces this improvements to the community here, so they shared their work, and this is ongoing. Conversation like how we can utilize this work and how we can commit this into open JDK. You know right now is: let's say, like analysis phase, we're testing the solution, but good point: a is design it for G 1 and it's it can work with Java 8 as well.

B

So you know because, like there is some kind of issue with Jakarta and you Java releases, because Jakarta I believe officially supports only Java 11. So we cannot use all this beauty. I mean improvements from the listen Belize's in Java, 12 and Java 13, but like there is a solution for now go version on door or dragon well or just do one and do periodical explicit, full garbage collection cycles.

B

Good, so that's in short, about heap usage. Now let me tell a little bit about the future like so what we want to do like and where we are moving, I mean as a community, but now the next step is Hebrew sizing like how to resize this xmx parameter on the fly without the need to restart Juvia.

B

It's not easy to be honest like so. It's really hard for, let's say like for Jiwon like to do it from scratch, but I mean like to do it. We need to rewrite garbage collector from from scratch, but there is a possible workaround when we just can meet much more RAM in advance and then specify okay. This is in you soft limit, so something similar was introduced. It inside GGC garbage collector, so the parameter is called like soft max heap size so, for example, how it works.

B

You specify like it's a mixed I'd say like 64, gigabyte right and then you say like okay, like soft soft max heap size is like 32, gigabyte and JVM will try to keep used ram under the 32 gigabyte of ram. So it will try to. You know like to call garbage collection cycle as more time as RAM usage comes close to this limit. However, if g vm just cannot live without extra allocation beyond this limit, soft max heap size, then at zg c implementation, z, GC, will allow you to go higher soft max heap size.

B

So it will give you from this extra pool you have.

B

This is the reason why it's called like soft so again, like just to summarize gvm will try to keep RAM usage below soft max heap size, but if it's like mandatory or the engine builder will give it higher, we will allow memory go high, as estimate good, we are working on the same, would say, solution form g1z a little bit different like we want to introduce like hard limit because, like we think, and we believe we must have a hard limit as well if Julian goes higher, because what is possible like it's, it's crazy, but it it can be done like so.

B

You can set its a mix higher then contain your max limit so because you can be sized containers limits on the fly without to restart of the container mount. So later you can go like you know, you can resize it, but you cannot resize it same mix, so you can resize on the softmax cube size right so and the idea is play with all these parameters.

B

Right, like so, you said, it's a mix hires and contain your max limit, and then you specify new max limit like soft max hip size, and then your GBM will work inside this space here.

B

So but if it's needed then you can in if I mean, if more RAM is needed, then you can increase, contain your much limit on the fly and increase this soft max heap size or hard max cube size on the fly and like and enjoy elasticity. You know adjusting maximum limit on the fly without downtime.

B

Something similar was introduced it and implemented in OpenGL. You can adjust max heap size through memory MMX bin. Actually it will be good to get some feedback from people who did it, who play it with the solution. But again, as you can see, community is moving into this direction and the most advanced, of course like was zing zing zing JVM from as all systems, because they is they created many years ago. I am not sure, but not many people I'm not sure. Why? But not many people knew about this and they have very unique solution.

B

It's fine! You can tune it in a different way. So then, when you're, actually in a good point, for example, you don't need to change the source code and configurations of your GBM, for example, if you just bring kazi's and it's so smart hey, it will not throw out of memory error or like omq. It will just give you extra RAM from common pool, so there is really like unique magic behind it, like it's out of scope of this presentation, but like it's possible to do now with C for garbage collector at Xena GBM.

B

So, just for your understanding, like you, can follow up after the conversation and if you are interested I can provide specific details how to configure it so guys this is, in short, I hope. It was clear that there.

C

Are a lot of material yep yep we've got some questions here them and there's a question from.

C

Man from Sablan that he talked about is LS DC based on memory. Only can you also have I know you're talking about only memory here you are talk a little bit more about elasticity on processors, disk and other things. Oh very.

B

Good, so look elasticity for processor, it's a little bit by default. Now it's available by default. There are no problems, so your GBM will not be killed if it requests more CPU, then it's available. So, of course you might have increased whatever is inside the operation operational system, but it's not clear, very critical, word say right. So that's the reason. Actually we want to make memory as elastic as CPU to you know like to get rid of this out of memory errors.

B

Yeah. That's one point then like talking now about disk space and network, they can be elastic as well. So today it depends on the service provider like if you go to, let's say to like large service providers who offer additional machines. You always pay for the limits for like how much you expect to consume at the maximum point. We believe it's not fair like right like so. This is the reason why we promote very much- let's say elasticity of sorry pay pay as you use.

B

So, for example, and this graphic you have like CPU memory, network and diskspace right, like you, can see like limits, they're much bigger than real usage like. Why should people pay a like for what they don't? Usually so many service providers they use a just over, sell it to other customers. Like you know, it's not fair like so it's better to come up with smart idea, like you know how to rebalance Lord inside the cluster and offer patrons model and big window sexually.

B

They started to adopt this model now several s solutions and they offered paper use model. So, to summarize, like.

C

I know you're talking about elasticity in terms you know in terms of memory and paper use. Can you can you tell a little kid? That's one of the first question we got here from.

C

Here, first guys, I got here from Vlade and Vlad. Ii was a little bit confused in the beginning, because you're talking about elasticity and we're talking about gel, ask here- and you didn't talk about your Laska at all so far right. So the whole discussion was about job in elasticity on Java. But can you tell us a little bit how how jelastic itself is using the elasticity of Java and what Julis is okay.

B

So because look like it was not about jelastic session, it was about Jakarta and OC city and so on, yeah. But if people interested effin it they can tell about jelastic so jelastic. First of all, it's a crony of java elastic right.

B

Let's say like seven years ago, like again, like I, was simple developer like and I faced the same problems like I was unhappy with so greedy Java, like you know and decided. Okay, like maybe we can solve this problem somehow, then they discover the kids. There are containers in the market and they elastic and like PHP is elastic and another kind of solutions are elastic, but Java is not elastic. So maybe we can solve this problem, so this is the reason why we started to.

B

You know like to put a force behind this moment and I started to talk to different people around to people who are working on garbage collectors more like to to get their opinion like what they think about this and honestly at 2011, I heard negative feedback, like other people.

B

Don't need it like so because, like you know, performance is number one now like, so people are fine to run as this, however, later when things came and became a put over like when more visual environments became a popular work, so the problem became more serious and more people started to pay attention this, like so, fortunately other people that develop different solutions. You know started to contribute like open JDK and they started to design new garbage collectors, keeping in mind elasticity from the beginning.

B

So again like talking about jelastic like jelastic, it's a platform as a service which was focusing on Java elasticity from the beginning, I say. However, it's a platform service for all languages. We are not religious about languages, we laughs, you know like oven, which is because it's matter of your business goals like so you can do some job faster with PHP and some job do better with Java. Like begin like, it depends on your task, like what kind of skills you hear like what.

B

What might say talents are available on your market like and so on, and so on. So jelastic it's a elastic platform as a service for developers. We again introduced city for java and we introduced it later use models. I would say that that's a two major contribution to the community yeah, but in general, again like in general, another good point like as a company. We are software companies, so we don't only the centers at all. Like so jelastic, it's a like a cloud solution.

B

There are many service providers around the world like we offer we evangelize freedom of choice like so. You can go and and shoot and check sorry service providers in Europe, for example, right and like in Latin America and like multi-region, for example right and if you don't like one of them like you, can easily migrate to another one. So it should not be allocating.

B

You should not be looking into one specific provider right, yep and actually I believe we should have competition like because like, if only like three big major players grabbed like the whole market like right like, and maybe it's not it's, not good for for the whole community. Like so I believe we should we should. You know we have more chance of the market getting shot about. Do a thing.

C

Okay, that's great so there's some more questions here, so glad stations. To sum up, is there either if there is a performance footprints? What would be the best use case should consider when an alien lasticity on JVM, so basically I mean. Do you need to do anything as a developer or do we need to do anything specifically or if we just use Java 12, that's enough very.

B

Good question I like it very much because it depends on your source code. If you write good source code, I mean it lists memory, so you have no memory leaks. Then you just use Java 12 and it's enough, so you will need to load a little bit like okay, like how the fine-tuning, but this is the reason why I'm here so I want to help you to understand like how it works like and maybe to get your feedback like you know about inconvenience.

B

So then they can bring back to people like anime can do some additional automation around it. So but the general, if you write a good code, so it will be enabled by default like so, and I saw in many cases when customers just come to us and they say like I like we face, may even like out of memory errors like we said the cookie just enable this option and your memory just disappeared. Like so I don't know it. Can it can help you even without code change, okay,.

C

All right there was a broad he saw some graphics.

C

Can we add a monitoring hook to send alerts in case of reaching out just images, Oh.

B

Reaching all the limits its you mean, like xmx limiter, or which one.

C

Is by no there's a lot of limits that you are like different plants right here.

B

Right, for example, so if you reach like it's, a mix, application will throw like or like. Then you can sign up, as there is specific option out of memory on out of memory error. Let me try to find it on.

B

Error, it should be like this I, don't remember exactly but yeah like you can specify it some specific option.

B

Yes yeah, so you can subscribe on this, it went, then you can do something like, like you know way, send send mail to. She sent me like guys like take a look. That is an exception but I but I believe the rest like you need to. You know analyzing box, so you can enable locks and check garbage collection cycles. All you need to write like eggs and mix bean to check what is going on inside or jar.

B

You have to check it from your application, because you can you can analyze the real usage genetically from your application, okay, yeah, by the way, if you go to the leans, it is an example. It is a source code of this vertical scaling test so and there is an example how I check memory usage yep, so you add memory MMX bin and then you get like hip memory like used memory, committed memory max memory, so you can use it inside your application and you know specify some triggers.

C

Okay, all right thing: thanks Ruslan, thanks Rulon, you know there is so much to learn about garbage collection. I've been ignited lots of new ideas. There was Edwin and the assays thanks Ruslan Vladdy, say: thanks was good into last season. Jdm.

A

C

There is any wage of clear unusual matter: space.

B

Very good question: ok, I didn't have to go this like I, can compulsory now so hi yeah, like I, don't take this question. I will do some homework like so but like I believe it should be elastic by default, so it should be elastic by default like so. But it's it's really like next topic to talk about like so. Let's fix its beginning hip because, like the majority of problems are inside the hip.

A

C

All right so guys, so everyone Tianna puts some of the links there on this on the on the shat right. So you can you know you can get more information. You know there's a little video. They can talk to more about this, there's also tweeter from jelastic, and we as soon as we finish here. We're gonna go on the tweeter right there, tweeter coms, last jelastic, we're gonna go there and you gonna share the presentation.

C

I shared the recording of this presentation and also the link for it is like so you know just follow juh last right now, you're gonna get it okay.

B

Thank you guys for joining for the questions. If you come up with additional questions, feel free to follow up through social media, so we can. We can discuss right.

C

And now so there's a lot of different tech talks. Sorry! So thanks thanks for the Eclipse Foundation for hosting us here. They do a lot of webcast like this, so you should follow the Eclipse Foundation to be notified of the next presentations and there's us actually link right there on the sheds with the recording from this presentation here. So you just copy that, so you don't miss it later.

B

Thank you. Nobody, okay,.

C