2023-07-11 09:57:14 casket-http is updated in the repository & running the various retroforth servers now

2023-07-11 09:57:33 https://forth.works/casket-report.txt is a summary of the changes

2023-07-11 10:58:56 excellent

2023-07-11 10:59:04 is it safe to link it now?

2023-07-11 10:59:58 crc: Did you consider doing this the C way? Where it's buffered in software so the "write one byte" word doesn't actually do anything until end of line or full buffer

2023-07-11 11:01:33 That is a nice performance gain

2023-07-11 11:26:37 drakonis: should be good to link to

2023-07-11 11:26:54 cool

2023-07-11 11:31:27 veltas: I hadn't considered that, but it does make sense. I'll look into adding support for buffering.

2023-07-11 11:37:57 The downside is you need to flush

2023-07-11 11:38:21 And specify when you don't want buffering

2023-07-11 11:40:30 I think it's legit to just leave it out as a choice of taste

2023-07-11 11:41:09 But I do think that things like flush/sync are just 'expected' from any performant I/O interface at this point

2023-07-11 12:23:45 i've posted it to lobsters

2023-07-11 12:44:43 peaked at 942 connections, it's still up and responding :)

2023-07-11 13:03:20 nice.

2023-07-11 13:03:39 time to put it in hn

2023-07-11 13:05:01 here we go

2023-07-11 13:35:53 that's hitting harder; peaked at 1421 connections (still at 241 in process)

2023-07-11 13:54:26 not holding up as well under the pressure from hn; load average is staying between 60-75, with about 260 connections

2023-07-11 13:58:57 Do you think there's room for making it tolerant of HN?

2023-07-11 13:59:35 hn hammered it very hard

2023-07-11 13:59:40 I mark this as a victory, I mean lots of 'enterprise' frameworks can't perform this well under HN attention

2023-07-11 14:00:01 it's better than before; last time the server was completely unresponsive for a

2023-07-11 14:00:10 lol even got dang replies

2023-07-11 14:00:37 small victories?

2023-07-11 14:00:38 about two hours; this time it's slow, but I'm able to watch the processes launch & end via top(1)

2023-07-11 14:01:06 the hn post has been up for almost an hour

2023-07-11 14:01:08 Is the logo a PNG?

2023-07-11 14:01:26 the hn post for 1 hour and 30 minutes

2023-07-11 14:01:30 lobsters

2023-07-11 14:01:33 rather

2023-07-11 14:01:39 I wonder if it would load faster as an SVG?

2023-07-11 14:02:21 maybe; it's a png, 16kb in size

2023-07-11 14:03:07 should i subject factor's website to this next?

2023-07-11 14:03:16 i couldnt submit the url previously

2023-07-11 14:03:58 crc: Does it send anything compressed?

2023-07-11 14:04:19 veltas: no

2023-07-11 14:04:45 Maybe that's worth doing, of course only if you pre-compress it, I'd expect compressing to make it slower on demand

2023-07-11 14:19:14 still being hammered hard innit

2023-07-11 14:24:27 yes; still at around 120 connections and a 30-40 load

2023-07-11 14:26:00 How do you decide when to end a connection early?

2023-07-11 14:26:19 Or is that handled by the OS?

2023-07-11 14:27:18 timelimit(1) is used to cap runtimes, and I have a shell script that kills processes that appear to be hung (runs every 6 minutes via cron)

2023-07-11 14:35:02 what's the specs for the server hosting the website?

2023-07-11 14:35:04 and which OS?

2023-07-11 14:38:20 Is there a process per connection or one process for all connections?

2023-07-11 14:41:43 crc: Do you know which font the logo was made with?

2023-07-11 14:41:53 (if it was made with a font)

2023-07-11 14:42:41 there's a lot of ways to make it scale better

2023-07-11 14:43:11 hm, its still at the top of the frontpage in hn

2023-07-11 14:45:12 It's a nanode instance on linode, running FreeBSD 13.2

2023-07-11 14:45:26 oh

2023-07-11 14:45:28 of course.

2023-07-11 14:45:55 that does explain a bit

2023-07-11 14:45:57 One process per connection, started by inetd

2023-07-11 14:45:59 nanodes only have a single cpu, right?

2023-07-11 14:46:13 Yes

2023-07-11 14:49:35 What's the overhead for each process?

2023-07-11 14:54:34 19mb ram and about 3-6% CPU based on output from top(1).

2023-07-11 14:55:17 I wonder how much of that is shared

2023-07-11 15:15:35 crc: a question someone asked on another channel, why is retro called retro

2023-07-11 15:16:10 or rather, why was it called retro in the first place when the previous author originally chose a name?

2023-07-11 15:18:15 From the manual, http://fossils.retroforth.org:8000/nga/file?name=doc/book/tech-notes/historical-papers&ci=tip has an old email on this

2023-07-11 15:22:06 The relevant part is: "I was telling a fellow musician about my ideas, how it would be cool to have a small OS that isn't bloated and unmanageable like Windows... go back to the 70's and resurrect a line of software that died out. He goes "hmm.. sounds kinda retro..""

2023-07-11 15:24:08 heh

2023-07-11 15:39:18 Still getting hammered, but this is the first time that it hasn't been so loaded down that I couldn't connect to the server and keep an eye on it.

2023-07-11 15:41:32 I should look at setting up a more capable server instance at some point.

2023-07-11 16:21:23 I think the free tier on Oracle is better than that, there are other 'free' providers as well if you don't want Oracle

2023-07-11 16:21:44 Opportunity to make some savings albeit small

2023-07-11 16:30:27 there are other good providers, yeah.

2023-07-11 16:30:42 i have a nanode but i'm not using it to host anything heavyweight

2023-07-11 17:44:54 impressive, its still in the front page

2023-07-11 19:29:04 veltas: re: font; I think it was derived from a predecessor to Bank Gothic or New Academy

2023-07-11 19:29:44 it's hard to recall for sure; the image was created somewhere between 2004-2007