Nice explaination man but one question, about last 3 mins of your video, Where you are creating virtual threads and platform thread & starting it. My question is, When you will start virtual thread then will it get map to platform thread ? If so then why directly mapping to platform thread is more time consuming ?
I have a confusion in your "high level benchmarking" around 11th min, of video.. Does No. of threads primarily depend on RAM side...I believe CPU CORE is the most prime factor to consider number of threads supported for a h/w system. Suppose you have 4 core CPU, i believe as per guidelines, we may have 4-5 threads at a time...Please share me more input in this regards.
Queries: 1. Why is CPU utilisation calculated as 16%? Can you explain that, why’re we taking it as a pct of 1sec. Is CPU blocked for whole 1 sec? 2. While showcasing whether virtual thread will release PT when blocked, you’re printing details for first VT only, sleeping for all 10. Why not just create 1 VT, and test?
Mega Star (Basant): Thank you for this information. I have a doubt. As virtual threads are residing in heap memory. if we increase the count of virtual threads then heap memory increases which leads to JVM performance issues. So what is the secret/logic/magic exists without having JVM performance issues . can you pls tell me in brief ?
Yes you are absolutely correct Virtual threads have a smaller memory footprint compared to traditional (OS) threads. However, each virtual thread still consumes some heap memory for its stack and associated objects. Increasing the number of virtual threads will increase the total heap memory usage. but still we can play with memory tuning to mitigate it partially like Ensuring virtual threads are not lingering unnecessarily. Properly manage the lifecycle of virtual threads to release resources as soon as they are no longer needed.also we can think about improving GC algo to make it better by profiling memory usages like JConsole or any other profiling tool .
IntStream.range(0, totalThread) .mapToObj( thCount -> Thread.ofVirtual().unstarted(() -> { // out.println("Virtual thread "+ thCount); })).toList(); without print statement: millis used to launch 10000 vthreads: 41ms millis used to launch 10000 pthreads: 735ms With print statement: millis used to launch 10000 vthreads: 1865ms millis used to launch 10000 pthreads: 813ms when print statement is there, the virtual thread took more time when compared to the platform thread. can you please explain why>
the result is so obvious if you want to handle 1000+ no. of requests with normal native threads you will have to spin up multiple instances of that particular application or scale up horizontally or vertically. But with virtual threads same target can be achieved using a low config instance. And cloud is not free it charges fee as Pay as per usage model. So with the usage of native threads be prepared for high amount of bills .
Why you used join() method for all virtual threads ? I tried doing running same snippet without join method it was not printing anything for 0th index , can anyone help ?
Maybe your main thread finished its execution even before the virtual thread finished its execution. You can either do join on virtual thread or put a forced sleep so your main thread is kept alive till the virtual thread is completing its execution.
Few questions 1. Why the 100ms which is actually processing the logic is not calculated during this process 2. If only 16% cpu is used then how are we getting memory out of bound exception
Great summarization of virtual threads. Can't wait any more for the upcoming videos on the same. Appreciate all your efforts in publishing this awesome content Many Thanks 😊
Total CPU time for single request = 0.1 (Request) + (0.1 Response) = 0.2 ms ??. Why was the IO (DB and other services ) 100 ms was omitted from Total CPU time for single request calculation ?? Am i missing anything basanth?
why will a network call will have any cpu usage ? . The cpu is not processing data at that time, its just waiting for response from DB or other microservice. Hence your cpu didn't got utilized properly. Whereas virtual thread ensures, when these scenario occurs platform thread is detached and used for other requests as a carrier thread. Improving scalability of app and maximizing the cpu usage.
Hey @Basant , Thanks for the video. One basic question, is it true that we can create any number of threads based on the Ram size ?, my understanding is the number of platform threads is equal to the number of cores in the machine we use. Is it not true ?
The short answer is yes, but you loose the profiling and observability of your application if you use reactive libraries also thinking reactive and writing asynchronous code is not so easy and exception handling is also pain. It takes time to understand since we're accustomed to writing synchronous style code. Structured concurrency is another JEP aims to solve this issue. All in all you get all the benefits of asynchronosity without writing asynchronous code. I hope this answers your question
Guys reactive and virtual thread both context is completely different please give me sometime i will help with your Confusion. Single statement it's not comparable
I love the Java Community Process and what it's doing to improve the Java language!
Thanks, Basant, for new future on Java 19 Virtual Thread architecture Appreciate your efforts.
Great Explanation sir, Jai Jagannath, May god bless you with good health.
Thanks for your videos. It helps a lot❤. If possible, try making a video or series on new features from JDK9 or JDK 11.
Great Explanation sir ♥
Awsome thank you sir ❤
Nice explaination man but one question, about last 3 mins of your video, Where you are creating virtual threads and platform thread & starting it.
My question is, When you will start virtual thread then will it get map to platform thread ? If so then why directly mapping to platform thread is more time consuming ?
Creating VT is not time consuming that's what I show in video last example
How many threads do you get per each spring boot tomcat server? How do you change this setting?
Thanks Guruji
Hi @Basant Sir ،
Kindly plan a multi threading programing playlist as there is no such proper playlist with real time use cases available on UA-cam.
Okay Sajid .
I have a confusion in your "high level benchmarking" around 11th min, of video.. Does No. of threads primarily depend on RAM side...I believe CPU CORE is the most prime factor to consider number of threads supported for a h/w system. Suppose you have 4 core CPU, i believe as per guidelines, we may have 4-5 threads at a time...Please share me more input in this regards.
How many default virtual threads can be there ? Any specific limit ?
Queries:
1. Why is CPU utilisation calculated as 16%? Can you explain that, why’re we taking it as a pct of 1sec. Is CPU blocked for whole 1 sec?
2. While showcasing whether virtual thread will release PT when blocked, you’re printing details for first VT only, sleeping for all 10. Why not just create 1 VT, and test?
Will you be able to upload this PPT or documentation in github?? Or in the README file??
is it like event loop implementation in other programming languages ? Thank You !!
Mega Star (Basant): Thank you for this information. I have a doubt. As virtual threads are residing in heap memory. if we increase the count of virtual threads then heap memory increases which leads to JVM performance issues. So what is the secret/logic/magic exists without having JVM performance issues . can you pls tell me in brief ?
Yes you are absolutely correct Virtual threads have a smaller memory footprint compared to traditional (OS) threads. However, each virtual thread still consumes some heap memory for its stack and associated objects. Increasing the number of virtual threads will increase the total heap memory usage.
but still we can play with memory tuning to mitigate it partially like
Ensuring virtual threads are not lingering unnecessarily. Properly manage the lifecycle of virtual threads to release resources as soon as they are no longer needed.also we can think about improving GC algo to make it better by profiling memory usages like JConsole or any other profiling tool .
What is the usage of threads in our applications😊
IntStream.range(0, totalThread)
.mapToObj(
thCount -> Thread.ofVirtual().unstarted(() -> {
// out.println("Virtual thread "+ thCount);
})).toList();
without print statement:
millis used to launch 10000 vthreads: 41ms
millis used to launch 10000 pthreads: 735ms
With print statement:
millis used to launch 10000 vthreads: 1865ms
millis used to launch 10000 pthreads: 813ms
when print statement is there, the virtual thread took more time when compared to the platform thread. can you please explain why>
When you use Prinln statement JVM use main thread which is nothing but platform..
How are we able to create more million of platform threads if they are mapped to OS thread which is dependent on number of Cores ?
Here thread uses context switching to get cpu time to run on top of OS thread. This is concurrency not parallelism.
To address the issue of thread per request, wbclient is also there then why virtual thread?
Buddy you might not be getting the concept properly could you please check again
In the world of cloud era, how best is this virtual thread impact? Can you summarize the benefits in view of cloud comparison..
Okay i will do some R&D with cloud and update you
the result is so obvious if you want to handle 1000+ no. of requests with normal native threads you will have to spin up multiple instances of that particular application or scale up horizontally or vertically. But with virtual threads same target can be achieved using a low config instance. And cloud is not free it charges fee as Pay as per usage model. So with the usage of native threads be prepared for high amount of bills .
Why you used join() method for all virtual threads ? I tried doing running same snippet without join method it was not printing anything for 0th index , can anyone help ?
Maybe your main thread finished its execution even before the virtual thread finished its execution. You can either do join on virtual thread or put a forced sleep so your main thread is kept alive till the virtual thread is completing its execution.
What happens to Thread Local ? Will the Virtual thread carry it?
Few questions
1. Why the 100ms which is actually processing the logic is not calculated during this process
2. If only 16% cpu is used then how are we getting memory out of bound exception
This is just a high level benchmarking buddy. The number could be very in the real system
Can we create pool of virtual threads like ThreadPoolTaskExecutor?
Yes we can and that's my next video using spring boot.
Great summarization of virtual threads. Can't wait any more for the upcoming videos on the same.
Appreciate all your efforts in publishing this awesome content
Many Thanks 😊
Total CPU time for single request = 0.1 (Request) + (0.1 Response) = 0.2 ms ??. Why was the IO (DB and other services ) 100 ms was omitted from Total CPU time for single request calculation ??
Am i missing anything basanth?
why will a network call will have any cpu usage ? . The cpu is not processing data at that time, its just waiting for response from DB or other microservice. Hence your cpu didn't got utilized properly. Whereas virtual thread ensures, when these scenario occurs platform thread is detached and used for other requests as a carrier thread. Improving scalability of app and maximizing the cpu usage.
@@DreamDropsTV thanks mayur
Hey @Basant , Thanks for the video. One basic question, is it true that we can create any number of threads based on the Ram size ?, my understanding is the number of platform threads is equal to the number of cores in the machine we use. Is it not true ?
Yes that's 100% true, even though you will create more threads concurrently only 4 threads will execute if Ram size is 4.
Thanks
Need more clarification, stil have doubt, this can be done in reactive programming
The short answer is yes, but you loose the profiling and observability of your application if you use reactive libraries also thinking reactive and writing asynchronous code is not so easy and exception handling is also pain. It takes time to understand since we're accustomed to writing synchronous style code.
Structured concurrency is another JEP aims to solve this issue.
All in all you get all the benefits of asynchronosity without writing asynchronous code.
I hope this answers your question
Reactive is hard to code and debug and restrictive.
@@sommyaruproy8405 we have desiged rate limiter using reactive, it was quite easy i guess. But might be hard in other use cases
Guys reactive and virtual thread both context is completely different please give me sometime i will help with your Confusion. Single statement it's not comparable
@@manikantareddy7595 I think reactive programming using PT thread ,then we can't achieve more request as come with visual thread
ThoughtPut - work done by cpu in a given time
But you should map a thread to a cpu core not ram
🙏👍
Thanks alot Sir..🙏