Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 4 years ago.
Improve this question
If a distributed computing framework spins up nodes for running Java/ Scala operations then it has to include the JVM in every container. E.g. every Map and Reduce step spawns its own JVM.
How does the efficiency of this instantiation compare to spinning up containers for languages like Python? Is it a question of milliseconds, few seconds, 30 seconds? Does this cost add up in frameworks like Kubernetes where you need to spin up many containers?
I've heard that, much like Alpine Linux is just a few MB, there are stripped down JVMs, but still, there must be a cost. Yet, Scala is the first class citizen in Spark and MR is written in Java.
Linux container technology uses layered filesystems so bigger container images don't generally have a ton of runtime overhead, though you do have to download the image the first time it is used on a node which can potentially add up on truly massive clusters. In general this is not usually a thing to worry about, aside from the well known issues of most JVMs being a bit slow to start up. Spark, however, does not spin up a new container for every operation as you describe. It creates a set of executor containers (pods) which are used for the whole Spark execution run.
Related
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 3 years ago.
Improve this question
I am writing a Java Application that aims to get and persist messages in real time. Beyond that, my application should care about several non-functional requirements like lineage control, transaction control, security, logging, monitoring, and so on.
Each feature is defined as a module and will be controlled within a thread concurring to my hardware resources (disk, memory, CPU, and GPU). As consequence, each time my application evolves, my thread control should evolves too.
To deal with that I am creating a global ExecutorService to manage all threads in my application. Some of these threads are permanent and defined as daemon. My application also control multiple sources. Each one representing a set of all features described above, evolving arbitrarily.
Which are the best practices for that scenery? How to control multiple threads while some will be created and dropped arbitrarily (as daemon or not) and another ones will be executed regularly nested or not to another threads?
Generally, if you do not want to manage creation and disposal of threads, use a thread pool (you can create one with ExecutorService executorService = Executors.newFixedThreadPool(10);).
However, be careful of premature optimisation (i.e. spending effort "optimising" code, at the expense of readability/maintainability, often when the performance gain is irrelevant). General good practice is:
Write clean, readable code.
Run your code and see if you encounter performance issues
Profile your code (e.g. with JProfiler or JVisualVM) to identify performance issues
Refactor your code if there is a significant problem with the way it is.
Remember, in general, developer time is more valuable than CPU time. Obviously there are limits to this, but usually you are best served keeping your code clean and readable.
You can also use this process to tweak parameters (e.g. on a fixed thread pool, there are various thresholds that can be fine-tuned to improve performance).
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 6 years ago.
Improve this question
I'm currently rewriting a Ruby on Rails web app in Spring Boot. A big part of the move is for performance.
Whilst developing the app, when I hit run in IntelliJ the first response time is typically around 1s which I assume is JVM startup, after a refresh it'll jump down to 300ms~ then 150ms for 4-5 further requests, after that it settles on 50-75ms for the most part. Randomly though later on I'll get a 150ms response again.
As a JVM novice I'm wondering what factors are at play here in the varying response times? which would be closer to the standard "hot" response times that I could expect in production? I realise I'm unlikely to get an accurate depiction of production performance on my local dev machine but would like to understand the variance seen above so I can at least gauge a little better what affect my incremental changes are having.
As a JVM novice I'm wondering what factors are at play here in the varying response times?
startup:
jit warmup
lazy initialization as part of your application
GC needing to settle on some heap size
steady state:
GC pauses
application behavior, e.g. cache entries expiring every now and then
varying load
JIT deoptimizations/recompilations due to some uncommon paths being taken
thermal CPU throttling, especially on but not exclusive to laptops
For server applications you should ignore the ramp-up behavior and focus on steady state. And guessing what the issue might be will not help, measurements are king.
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 8 years ago.
Improve this question
I was wondering if a simple program with no threads can run faster on some computers which have many cores? or on a cluster of linux servers?
Recently I have run my algorithm which has to process billions of IP packets on my PC(core i7 with 16GB RAM) and it took 1881 minutes to finish processing. Then I thought its good to run the algorithm on clusters of linux servers each node with 10 processors and 48GB RAM to get the results quicker. However, there is no big difference between the two experiments.
Can someone comments what I am missing?
Unless your algorithm actually makes use of those multiple instances and extra memory, there shouldn't be a lot of difference. Parallel programming is an art of its own, and a "regular", single-threaded program doesn't just change into parallel one by itself.
If you have a single thread of execution more cores, CPU or machines won't help. Only a faster CPU would speed things up, and that only if your process is CPU-bound, and not IO-bound.
First you should check, where your processing time is spent, in CPU, or waiting for IO. If you have a significant amount of CPU usage, you can try to parallelize your work, i.e. split the data into chunks, and have different thread resp. machines process them in parallel.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 8 years ago.
Improve this question
Or will it be possible in the near future? Can they optimize JVM further and implement even gentler garbage collectors, maybe with some support for lower-level interference?
I could have asked the question if it's possible to develop AAA games, but since AAA games use highly advanced game engines, I thought the barrier to overcome here is to have similarly advanced game engines written to the JVM.
Is it likely that in 5 years we will see games like Assassin's Creed, that are optimized to infinity with the help of C/C++ libraries by necessity, written in Java?
For the most part, the intensive work is done on the GPU and these are getting more powerful all the time.
You can use the GPU from Java already and JavaFX does this as well as a number of CV libraries.
A simple way to get gentiler garbage pauses is to create less garbage. You program will run faster as well.
Another effective technique is to move the bulk of your data off heap. For minecraft server this can reduce the heap size by 80% and significantly reduce the frequency and pause time of collections.
http://vanillajava.blogspot.co.uk/2014/06/minecraft-and-off-heap-memory.html?m=1
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 9 years ago.
Improve this question
I've been writing some console applications in c++ for working with audio for a little while now and I'm interested in running them on a website. Most of my programs are quite resource-hungry, however, some with execution times of up to 5-10 min, reading and writing several gigabytes to and from disk, and requiring several gigabytes of memory. I've done a few simple php-mysql pages before, but nothing like this, so before i get my hopes up and dive into learning how to get an application running on a website, i figure i should ask a few questions:
Is it even feasible to run a program like this on the web? How would performance on a server compare to my PC?
Do web hosts typically allow a single user to use this kind of memory?
I realize c++ isn't usually the first choice for web programming, but since performance will be critical would it be better than Java?
I know nothing about this, so i'm just trying to get my expectations straight.
This is my opinion:
1 - The user of your web application is probably not going to wait 5-10 min for a response. You can focus on doing the hard-work on another process and your web app later shows the results to your user in some way.
2 - Yes, they allow, but that costs money. You can see Amazon EC2 and Digital Ocean (cheaper).
3 - The programming language in this case (C++ or Java) is not that important. Focus more on your problem, architecture, deferred tasks, batch processing, etc. That will really make a difference.
No, the programming language doesn't much matter. It used to be the case that java was slower than C++ i believe, but that gap has closed pretty much as compilers have improved. If you want to run your applications better, try to design them in such a way that they are very efficient. Looking into Time Complexity may help, if you haven't already done so. The better your time complexity, the faster your program.