Hi guys, I'm wondering whether Pharo support any form of non-blocking IO. You can find such IO support in Node.js or Java NIO packages and on Linux they use select/poll async IO system calls. An example scenario is to be able to schedule 10 concurrent HTTP calls taking 1-5s and wait for them to finish. Naturally, if you perform the calls sequentially you'll wait for much longer than if you fire the calls concurrently. Is it possible in Pharo? I've only spent a few hours reading about the platform but I'm still not sure whether the built-in process/threading supports such scenario. -- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
I think you can do what you describe if you use #fork to create separate processes for each of your concurrent HTTP calls. Here's an example -- Try the following (from http://forum.world.st/Teaching-fork-td4786444.html) 10 timesRepeat: [ (Delay forSeconds: 1) wait. Transcript show: 'hello'; cr ]. Then the following : [ 10 timesRepeat: [ (Delay forSeconds: 1) wait. Transcript show: 'hello'; cr ]] fork. You can also spawn OS sub-processes: https://github.com/pharo-contributions/OSSubprocess -t -- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
Thanks. Are there any benchmarks available for Pharo HTTP servers? Is it possible to achieve a performance similar to node.js? I know the biggest selling point of Smalltalk is not perf but still I'm curious. The Concurrency book [1] says that Pharo's Process is like a fibre and is lightweight. How big is the stack? Java Quasar's fibers are very lightweight with a 400 byte stacks [2]. Can you have hundreds or thousands of processes running blocking IO without much memory and performance penalty? [1] http://books.pharo.org/booklet-ConcurrentProgramming/ [2] https://docs.paralleluniverse.co/quasar/javadoc/co/paralleluniverse/fibers/F... -- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
Daniel,
On 24 Apr 2020, at 12:35, Daniel TurczaÅski <mail@danielturczanski.com> wrote:
Thanks. Are there any benchmarks available for Pharo HTTP servers? Is it possible to achieve a performance similar to node.js? I know the biggest selling point of Smalltalk is not perf but still I'm curious.
Benchmarking is a black art with many pitfalls. Let's say that Pharo HTTP performance is acceptable. Here is a test that I just ran (on a LXC/LXD container on a big production server). t3@t3-pharo-test:~/tmp/pharo9$ cat /etc/lsb-release DISTRIB_ID=Ubuntu DISTRIB_RELEASE=18.04 DISTRIB_CODENAME=bionic DISTRIB_DESCRIPTION="Ubuntu 18.04.4 LTS" t3@t3-pharo-test:~/tmp/pharo9$ curl get.pharo.org/64/90+vm | bash % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 3054 100 3054 0 0 198k 0 --:--:-- --:--:-- --:--:-- 198k Downloading the latest 90 Image: http://files.pharo.org/get-files/90/pharo64.zip Pharo.image Downloading the latest pharoVM: http://files.pharo.org/get-files/90/pharo64-linux-stable.zip pharo-vm/pharo Creating starter scripts pharo and pharo-ui t3@t3-pharo-test:~/tmp/pharo9$ ./pharo Pharo.image printVersion [version] 'Pharo9.0.0' 'Pharo-9.0.0+build.252.sha.fe2815c89f309789cbfba9a1ce821fb959dfe987 (64 Bit)' t3@t3-pharo-test:~/tmp/pharo9$ ./pharo Pharo.image eval --no-quit 'ZnServer startDefaultOn: 1701' & [1] 32203 a ZnManagingMultiThreadedServer(running 1701) t3@t3-pharo-test:~/tmp/pharo9$ curl http://localhost:1701/random 0BFE44ECF0BEEDDD949A8D5A363E5E6192A4B7CD4939E8BD7A7C7EC69995157 t3@t3-pharo-test:~/tmp/pharo9$ ab -k -n 1024 -c 8 http://localhost:1701/random This is ApacheBench, Version 2.3 <$Revision: 1807734 $> Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/ Licensed to The Apache Software Foundation, http://www.apache.org/ Benchmarking localhost (be patient) Completed 102 requests Completed 204 requests Completed 306 requests Completed 408 requests Completed 510 requests Completed 612 requests Completed 714 requests Completed 816 requests Completed 918 requests Completed 1020 requests Finished 1024 requests Server Software: Zinc Server Hostname: localhost Server Port: 1701 Document Path: /random Document Length: 64 bytes Concurrency Level: 8 Time taken for tests: 0.246 seconds Complete requests: 1024 Failed requests: 0 Keep-Alive requests: 1024 Total transferred: 256000 bytes HTML transferred: 65536 bytes Requests per second: 4155.22 [#/sec] (mean) Time per request: 1.925 [ms] (mean) Time per request: 0.241 [ms] (mean, across all concurrent requests) Transfer rate: 1014.46 [Kbytes/sec] received This is a small, do nothing request, tested locally, and so on. I would say it is safe to say you can handle 100s to 1000s of requests per second on a single Pharo instance (which is using only a single core). The good thing about HTTP is that you can normally scale horizontally by adding more and more worker instances under a load balancer -- these are standard techniques. HTH, Sven
The Concurrency book [1] says that Pharo's Process is like a fibre and is lightweight. How big is the stack? Java Quasar's fibers are very lightweight with a 400 byte stacks [2]. Can you have hundreds or thousands of processes running blocking IO without much memory and performance penalty?
[1] http://books.pharo.org/booklet-ConcurrentProgramming/ [2] https://docs.paralleluniverse.co/quasar/javadoc/co/paralleluniverse/fibers/F...
-- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
Thanks Sven! Looks good to me. -- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
Asynchronous I/O and non-blocking I/O are very different things. The POSIX aio* functions are asynchronous, not non-blocking. The "conventional" Unix way to do asynchronous I/O is to start a new thread for the transfer. The new thread uses ordinary synchronous I/O and then responds to completion any way you want. It is a rather easier model to program for than asynchronous I/O. On Mon, 20 Apr 2020 at 04:15, dturczanski <mail@danielturczanski.com> wrote:
Hi guys,
I'm wondering whether Pharo support any form of non-blocking IO. You can find such IO support in Node.js or Java NIO packages and on Linux they use select/poll async IO system calls.
An example scenario is to be able to schedule 10 concurrent HTTP calls taking 1-5s and wait for them to finish. Naturally, if you perform the calls sequentially you'll wait for much longer than if you fire the calls concurrently.
Is it possible in Pharo? I've only spent a few hours reading about the platform but I'm still not sure whether the built-in process/threading supports such scenario.
-- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
Thanks for pointing out the difference between non-blocking and async IO. I think in Java NIO there is a thread that constantly calls the [e]poll() in the background and this way the IO becomes non-blocking. -- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
participants (5)
-
Daniel TurczaÅski -
dturczanski
-
Richard O'Keefe -
Sven Van Caekenberghe -
tbrunz