feasibility of a web forum in Python

[Skaperen]

what is your opinion on the feasibility of a web forum written in Python? i notice that the vast majority are written in PHP. even Perl has a limited showing for this category of application.

i've been thinking about a design where there is a client/server relationship with the server implemented entirely in Python (not counting things like the database, if used). the client might be done in JavaScript or one of the web client side Python implementations or even as a standalone app not based on a web browser. the important idea of this design is to make a clean and simple protocol between client and server so that either side is pluggable as desired.

i really like the look and feel of Discourse. the protocol should include information fields, or otherwise be flexible enough, to allow client side software to implement something like that.

your thoughts?

[Larz60+]

It can be done using Flask

[Skaperen]

yeah, Flask could do its networking layer, and manage a million clients spread over a dozen hosts.

[Larz60+]

Just an FYI the pinterest site is done with a combination of Flask, Bootstrap, Jinja2, and flask-sqlalchemy and services several billion accesses per day.

My site is done using the same tools.

Miguel Grinberg's new and improved Flask Mega Tutorial covers everything needed to be successful with such a task
https://courses.miguelgrinberg.com/

[snippsat]

what is your opinion on the feasibility of a web forum written in Python?

There is FlaskBB that's probably the most updated forum software made in Python(steadily worked on for 5-years).
The changes that we would move to it is very small,it still need more developing.

i really like the look and feel of Discourse

I did run NodeBB for a while(1-year) which is similar to Discourse,when we did choice this forum.

Now have we done lot of work on the MyBB version we run now,and it's now a very stable platform.
I am even glad that MyBB 2.0 has been put on hold indefinitely,as it would break a lot of stuff we have now.

We are not a small forum anymore and have to be careful not breaking stuff,just to try out other new stuff.
On the server side we have upgraded to 3-CPU,just to cope with traffic we have now.
Over 20000 user a day and 100000 a week.

[Skaperen]

Just an FYI the pinterest site is done with a combination of Flask, Bootstrap, Jinja2, and flask-sqlalchemy and services several billion accesses per day.

My site is done using the same tools.

Miguel Grinberg's new and improved Flask Mega Tutorial covers everything needed to be successful with such a task
https://courses.miguelgrinberg.com/

on how many machines? how much could one single 3.2 GHz 8 core server do?

---

i'm not proposing that python-forum.io be changed. if i mention it, it's for comparison, only.

[Larz60+]

on how many machines?

I haven't got a clue, more than 1 probably a lot more, or perhaps on Jan Grey's 1000 RISC-V cores on a Xlinx FPGA!

Actually, I see he's got a new one with 1680 cores, see: https://forums.xilinx.com/t5/Xcell-Daily...a-p/789549

[Skaperen]

i wonder how many memory channels it has.

---

"many" DRAM memory channels. i wonder how wide.

1680 cores at 375 MHz. that would like 225 cores at 2.8 GHz. i wonder how much these little cores can do. probably enough for the intended use. 8 core clusters, so there must be 210 clusters. that illustration did have 210 of something. 8 x 375 MHz = 3 GHz. of course that's not true equivalence.

[Larz60+]

There was an ACM lecture by Dave Patterson: https://www.youtube.com/watch?v=aA5pqklkkvI
which is very informative about this and other on going projects.
I was online for the original webcast, and it's a was good take.
worth watching

[Skaperen]

i dropped out of that video at the 2nd question because i could not hear the questions well enough. they should have at lest put the questions in text in later editing.

this did remind me of a little project i did back on the S/360 in assembly language. it had a16-bit multiply instruction called MH. i took a different method of multiplying that i came up with back when i was in high school and wrote a series of instructions to implement it. i had originally implemented it on the 6502 microprocessor (no multiply instruction as this was a micrcodeless ISA) at 8-bit scale, and it worked very well. so i scaled it up to 16-bit for the S/360 ISA (backed by microcode). this required a table of 512KB (in 8-bit only 1Kb). the sequence of instructions i wrote to do this took 75% of the time as the MH instruction. scaling to 32-bit would have been impossible back then as the table size would be 64GB. don't even think of going to 64-bit. for reduced precision integer multiply, this might have been a slight improvement.

edit:
the table size in 8-bit bytes would be: 2**(b+2)*(b/8)

that was a great video.