hyrcanus | how many GB drive space does GCC need to compile | 02:06 |
---|---|---|
Hydragyrum | as in compiling gcc? I don't know the exact number off the top of my head, but 4gb should be plenty for just gc | 03:23 |
Hydragyrum | s/gc/gcc/ | 03:23 |
hyrcanus_ | yeah seems i need more for gcc-10. freed up 7.8GB | 03:24 |
hyrcanus_ | there's a nasty bug with 'atomic' on my aarch64 board. a bunch of stuff won't compile. | 03:24 |
hyrcanus_ | and downgrading to gcc-9 seems to be problematic | 03:24 |
Hydragyrum | guess I misremembered that -- I know LFS previously had 10gb total space req | 03:25 |
Hydragyrum | aarch64's reqs are also probably different from amd64 which is what I've compiled it on | 03:25 |
hyrcanus_ | ARM is still a stepchild in many ways | 03:26 |
hyrcanus_ | on redhat's gcc with this armv8 defining -mno-outline-atomics is a workaround, but not on debian's | 03:29 |
hyrcanus | woke up, gcc compilation erroring, 7.8 GB used | 09:59 |
gnarface | hyrcanus: try using zram for swap | 10:31 |
hyrcanus | hm 3.7GB RAM, 6GB normal swap. | 10:32 |
hyrcanus | interesting idea | 10:33 |
gnarface | 7.8 may seem like alot but i recall hearing of a efl build bug that also only affected aarch64 which caused it to use around 32GB | 10:34 |
gnarface | but zram is like downloading more ram | 10:34 |
gnarface | so you could easily manage that | 10:34 |
gnarface | if nothing else you could at least find out how much more ram you can make it use... you know for science ;) | 10:35 |
hyrcanus | is the performance impact of zram significant? | 10:37 |
gnarface | not compared to actual swap | 10:37 |
gnarface | not compared to disk i/o | 10:37 |
hyrcanus | in the case of not-swapping | 10:37 |
gnarface | it's actually rather fast on modern hardware, and is a significant performance improvement over a physical swap partition for anything ram constrained | 10:38 |
gnarface | for anything not ram constrained you wouldn't be using swap so it wouldn't make a difference | 10:38 |
gnarface | the idea here is to make a zram block device then put your primary swap partition on that instead of the harddrive | 10:38 |
gnarface | then you can compress it at a factor of 10 or whatever and get as much ram as you need | 10:39 |
gnarface | lots of ARM devices only have 2GB of ram and they're not expandable, so this trick has been rediscovered in that community | 10:39 |
gnarface | (you can imagine the swap i/o delay is especially bad using a microSD as your main drive) | 10:40 |
hyrcanus | yes | 10:41 |
gnarface | definition of modern hardware in this context: anything after the pentium 1 | 10:41 |
hyrcanus | i have one arm device with zram enabled so at least i have a reference | 10:41 |
hyrcanus | 'zram-config' has no installation candidate | 10:43 |
hyrcanus | ok that's ubuntu-only | 10:43 |
gnarface | hyrcanus: it's zram-tools | 10:48 |
gnarface | probably | 10:48 |
hyrcanus | so for a system with 3754 mIB ram, echo 7G > /sys/block/zram0/disksize would be reasonable? | 10:49 |
gnarface | i don't know really but that sounds fairly conservative from what i've heard other people doing | 10:50 |
gnarface | people have reported easily using it for double or triple that | 10:50 |
gnarface | i'm sure the performance will scale inversely but i couldn't tell you where the sweet spot is | 10:51 |
gnarface | you can have more than one swap partition so it should be possible to keep the one on the harddrive but that could shackle your speed too | 10:52 |
gnarface | but speed isn't really the goal here i don't think, is it? | 10:52 |
gnarface | it's more about getting more space at any cost | 10:52 |
gnarface | i would think | 10:52 |
hyrcanus | i don't have a log of ram usage from last night's compile attempt | 10:53 |
hyrcanus | i can hear when my SBC is working. there's some resonance in the power supply :) | 10:57 |
hyrcanus | well now i seem to have lots of swap without a swapfile \o/ | 10:58 |
hyrcanus | other device has zram set up to 25% of physical RAM | 11:09 |
hyrcanus | sets up one zram device per cpu core | 11:12 |
alv | .... | 14:55 |
gnarface | yes? | 14:57 |
blockhead | Just got up, was reading the scrollback from the overnight and ... I had no idea zram worked that way. Fascinating. Thank you gnarface. | 15:00 |
gnarface | no problem | 15:01 |
gnarface | yea it's an idea i remember cooking up myself even as a child, but had discarded as i'm sure we all did because of the perception of how ludicrous the cpu overhead would be for such a task | 15:05 |
gnarface | but i guess the era where that was true has passed us by along with my childhood | 15:06 |
gnarface | once we hit gigahertz and gigabytes this type of compression became "trivial" | 15:06 |
* blockhead thinks of "ram doubler" on the mac in the mid 1990s | 15:07 | |
gnarface | but on uncompressed plain-text content, which gigabytes of ram surprisingly often is, can get nearly a 90% size cut | 15:07 |
gnarface | so it basically just uses your free space like ram doubler did, but as a separate block device | 15:07 |
* blockhead is reading the wikipedia page, it says pretty much what you did. Neat. | 15:07 | |
gnarface | (which you can do whatever you want with but for stuff other than swap, tmpfs might be a better choice) | 15:07 |
gnarface | it is reportedly surprisingly useful for improving firefox performance on systems capped at 2GB of ram | 15:09 |
fluffywolf | I was playing with a pmmx-233 a few months ago, with 160GB ram. I tried zram to see if it would make firefox run better. it made it much, much slower. | 15:43 |
gnarface | hehe | 15:43 |
fluffywolf | 160MB | 15:44 |
* fluffywolf is too used to typing ram sizes on modern systems. lol | 15:45 | |
fluffywolf | it went from spending its entire time swapping to spending its entire time compressing and decompressing, which was much slower on that cpu. | 15:45 |
gnarface | brutal | 15:47 |
fluffywolf | I gave up running firefox on it. | 15:48 |
fluffywolf | it's kinda sad how much ram firefox needs to load one tab. | 15:49 |
jushur | zram will be fast on a fast CPU obviously, also note that modern CPU's usually has dedicated paths for handling this type of software | 19:49 |
gnarface | anything with multiple cores over 1ghz and ddr2 or better ram should do well i'd think | 19:50 |
gnarface | ram speed counts for this too | 19:50 |
jushur | yeah | 19:50 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!