up_the_irons: mhoran: the only reason 40 Gbps Ethernet became a thing was because the backplane in Cisco switches couldn't go higher than that. And aren't they a big player in IEEE ? So it was codified and everyone was like, " wtf fuck you".. because they held back adoption of 100 Gbps ethernet
mercutio: that Arista... it makes no sense to have 40x 100 Gbps but only 8x 40 Gbps.. the uplinks are a bottleneck
brycec: looks like I can ping one of your IPs now (yay!)
mercutio: up_the_irons: that's why i thought it might be a typo it just seemed odd
but the datasheet matched
brycec: up_the_irons: What a wonder when systemd actually starts!
-: brycec moves on to figuring out what the hell happened that Debian forgot how to /sbin/init
brycec: (much smoother now that I can SSH in)
acf_: that's always my worst fear after my arp box has been running for 2 years
brycec: 6 months in my case
Upside: zfs snapshots are kept for 6 months
acf_: I was shocked everything started back up without any intervention after the cage move the other day
brycec: You lucky duck :P
well well well... The March 1st snap is good, the April 1st snap has a busted /sbin/init
-: brycec hugs zfs
up_the_irons: wow how did /sbin/init get corrupted
hack?
mercutio: maybe the data sheet is wrong? hahha
brycec: Went from being a symlink to some old-ass sysvinit
(narrowed it down to sometime between March 1st and March 25th... because I only keep 30 daily snaps)
acf_: you think the result of an upgrade?
brycec: Must be?
(No evidence of unauthorized access)
acf_: that system must have been installed a long time ago if it has any sysvinit binaries anywhere near it :P
brycec: I mean, it's Debian, so...
up_the_irons: heh
acf_: bit of a change of subject but
has ARP ever thought about buying ARM server hardware?
pretty niche atm, but aws is offering it now
brycec: Found a forum post (two, in fact) that described what happened precisely o/ https://forum.proxmox.com/threads/upgrade-to-kernel-4-15-18-12-pve-goes-boom.52564/#post-243333
acf_: might be cool to get in ahead
brycec: so it was zfs's fault in the end :P
(sort of)
brycec: shuddup :P
-: brycec blames packages
brycec: *packagers
mercutio: i wonder if the idea of the 8 uplinks is to connect to legacy 40gigabit networks for your 100 gigabit supercomputer cluster up_the_irons ?
up_the_irons: "pve goes boom" hahahah
acf_: never thought much about it... i'm so tired of buying hardware
it's the opposite of an investment... if you want to make sure your money runs the other way of inflation, buy technology
acf_: I suppose it is a necessary evil though
buying hardware
https://www.gigabyte.com/ARM-Server/R150-T62-rev-10
brycec: I wouldn't say buying is "necessary" - there's always theft :P
acf_: this looks like it's roughly US$5,000
or, more legally, leasing :P
cool thing about these arm boxes is they seem to have a ton of cores
so I presume you can dedicate 1 core to each of 47 customer VMs
on a single blade
up_the_irons: I've learned that to survive in this biz, you have to let someone else eat the depreciation. so, after AWS or Google offload 10,000 of these boxes onto eBay, then we'll buy some haha
acf_: hahaha fair enough
that's how I buy my laptops actually
always off ebay, and always at least 2 generations old
price is like 1/10th and the box is good as new
up_the_irons: yup
I'm still using my Lenovo T520 which I got for $240
I had another T520, which a client had bought me years earlier, and it was $1800 new
acf_: ah damn that's good
I was using an ebay X250 until recently
up_the_irons: nice
acf_: I decided I needed 16GB for all my chromium tabs so I upgraded to a 5th gen X1 carbon, also off ebay
brycec: Not to mention by waiting, there's either an established market for $newtech or you know it's a flop and not to waste your money :)
acf_: spoiler: chromium figured out how to eat the 16gb
mercutio: chrome is such a huge memory hog
i've started using firefox sometimes now
it seems to be getting worse and worse
computer cpus have hardly been rising in speed
they've actually been getting slower over time with all of these spectre/meltdown workarounds
acf_: https://unixcube.org/who/acf/tmp/images/screen/top-1.png
this pretty much describes it :P
mercutio: there's very little performance boost from ddr3 to ddr4
graphics cards have improved
acf_: chromium chromium chromium chromium chromium java chromium
mercutio: nvme is stupidly fast for disk
but single threaded processing speed struggles to make much improvement
up_the_irons: acf_: I have a tab addiction too
mercutio: and most code still isn't designed to parallelise well
the main benefit to performance ordinary users get now is things like JIT compiling of javascript
up_the_irons: I think it's time I leave the DC... it's been another like 12 hour day here
mercutio: if people coded for performance like they did in amiga days computers would probably rocket along
acf_: yikes, yeah time to go home :P
up_the_irons: :)
acf_: must be working quite hard with all this migration stuff recently
up_the_irons: mercutio: haha yeah
mercutio: acf: there's a lot of little things
up_the_irons: acf_: certainly, it's been a whirlwind
mercutio: but things are becoming simpler
up_the_irons: You basically start over, but with a fresh design
mercutio: ipv6 and ipv4 terminate on the same router
up_the_irons: and you take what you learned over the years and make changes
acf_: yeah all that plus the cage move
up_the_irons: our setup is soooooooo much simpler now; not quite as redundant, but the areas where we built redundancy before, in over 10 years, never needed it
our s1.lax still has over 10 years uptime
mercutio: also we have spares so we can still resolve any issues quick
up_the_irons: s1.lax uptime is 10 years, 9 weeks, 4 days, 13 hours, 10 minutes
acf_: 10 years is the standard service life of industrial electronics
up_the_irons: it will go down, not due to failure, but because we simply don't need it anymore
I'm going to keep it in my living room for a while
hhahaah
mercutio: and because it's physically huge
up_the_irons: I'll rest my feet on it
brycec: G'night #arpnetworks thanks for your help tonight. My server is back to its normal working order.
mercutio: it's also 10 years of doing a lot of changes to config
with no flash failures or anything
up_the_irons: brycec: cool!! great to hear!
mercutio: yeah.. not even a RAM failure
acf_: s3 on the other hand
mercutio: s3 is gone
up_the_irons: s3.lax was just a software router (OpenBSD)
and yeah, s3 is dead
acf_: oh s3 was the software router
for some reason I never made that connection
up_the_irons: yeah
acf_: iirc it was the one that always kept crashing :P
up_the_irons: s3, then s32, then r2
mercutio: we have hardware forwarding of ipv6 now
but r2 performance was fine
up_the_irons: acf_: b/c of Any2 IX terminating on it? That was actually s7.lax, a Cisco 6506. But it never crashed after we took Any2 off it
mercutio: it's just we get to consolidate
up_the_irons: yeah
acf_: ohh I must be thinking of s7 then
up_the_irons: yeah it must have had some bug
mercutio: that was a long time ago
acf_: haha feels like yesterday
up_the_irons: but it's done pretty well recently
s7.lax uptime is 4 years, 36 weeks, 3 days, 23 hours, 49 minutes
acf_: see how time flies?
acf_: especially when you haven't been on the irc in 3 years :P
mercutio: yeah irc isn't as active as it used to be
up_the_irons: alright guys, I'm gonna head out too... too much data center noise for today
well the +r really reduced user count
but we had to do it b/c of the spam
-: up_the_irons packs up
mercutio: well even before that it was pretty quiet
acf_: yeah I remember going on here for a minute during the supermicro thing
irc logs had been basically empty for days
mercutio: supermicro things?
acf_: "the big hack"
mercutio: oh right
acf_: seems like it's picked back up a bit the last week or so with all the stuff going on though
mercutio: a little
acf_: https://danluu.com/input-lag/
re what you were saying earlier
you've seen this yeah?
mercutio: have you done the reaction speed test?
https://www.humanbenchmark.com/tests/reactiontime
what i found most interesting was that my reaction time would get slower when i was tired, but i didn't feel like my reaction time was slower
acf_: looks like I'm at roughly 300ms right now
mercutio: it's kind of like how they say that driving sleep deprived can be as bad as driving drunk
i'm usally aruond 220
i'm like 280 when tired
mouse can make a difference though
some add like 20 msec latency
acf_: thinkpad integrated trackpoint for me
mercutio: 249 now
oh and screen can make a difference too
but if you use the same device and test yourself against yourself you can see how much you vary
oh also reaction times tend to be quicker when you're younger
acf_: lol I get like 500ms on my phone
mercutio: yeah phones are notorious for lag
it's all accumulative also
so there's the input device latency plus the screen latency plus the os/software latency
also you can speed your reaction up slightly by turning your head the side slightly and using peripheral vision
acf_: hmm my older thinkpad I get 230ms
mercutio: yeh device makes a difference
so 70msec changing device? :)
so like if you ssh'ed half way across the world you might notice a difference on your slower device vs your faster
but if you ssh'ed to a computer in the same city as you you probably won't
acf_: yeah that's a lot of latency
mercutio: i think if you play first person shooter games as a kid you'll tend to have faster reaction times too. but at the same time if you have slow reaction times you probably hate first person shooters.
acf_: hahaha yeah I've never played fps
mercutio: probably any twitch type games would help it
acf_: ever played vvvvvv?
mercutio: or sporting things that use reaction time
nope
acf_: https://upload.wikimedia.org/wikipedia/commons/5/52/VVVVVV_-_The_Tomb_of_Mad_Carew.png
only game that comes to mind I've played that requires good reaction times
mercutio: i've been playing factorio recently
which isn't really reaction time game
oh yeah old games used to be so hard!
and there would be things like you die once and you go back to the start of the game
acf_: actually that's my favourite thing about vvvvvv
that little "C" there is a checkpoint
they're quite often
mercutio: ahh
acf_: you only go back to that, and there is no concept of lives
mercutio: yaeh i think checkpoints were nice tbh
acf_: so you can just keep trying until you make it
otherwise I'd give up after 5 minutes :P
mercutio: i think games used to be too hard
but now they're too easy instead generally
acf_: can't stand re-doing a whole level because I hit a spike
mercutio: yeh
that was so common though!
acf_: I guess it instils discipline
mercutio: well also less code
acf_: haha really
mercutio: well things like sega master system had kilobytes of code i think?
like 20k?
acf_: yeah back in the day when video games were written in assembly
mercutio: well sega soem games were bigger than others
it has 8kb ram 16kb vram
so yeah i wouldn't be surprised if games were 20k
acf_: 20k including graphics?
idk the code and graphics would probably be stored on rom right?
mercutio: oh they went up to 1024kb
yeah so the roms can be quite big
surprising
acf_: yeah small ram makes sense
mercutio: quite a few old amiga games didn't use the whole floppy
like pre native games
acf_: the performance vs laziness (or "work") tradeoff you mentioned earlier is pretty interesting
mercutio: hmm the sega master system came out at $200
that's cheaper than i thought
i wonder what equivilent that is now
yeah around 2001 or so languages liek perl, pyhton etc were taking off a lot
using lots of ram and having subpar performance
acf_: java is what really comes to mind
mercutio: heh
yeah
acf_: when you say lots of ram and subpar performance
mercutio: so java when it came out was really slow
acf_: but that would have started in the 90s yeah?
mercutio: so android is java
yeah in the 90s
solaris used to be called slowaris
it was always a little sluggish
linux had a reputation for being fast and unstable
solaris for being slow and reliable
but lots of old solaris programs would fail pretty badly at times
acf_: java's awful performance seems to have kept up with improvements in computing power somehow
mercutio: yeah there's been so many new versions
acf_: eclipse is still super slugish and uses all my ram
mercutio: and it still seems to be kind of sluggish. also the amonut of ram programs use these days is insane
ios vs android is actually a good example of compiled vs jit code
acf_: yeah well some people thought it would be a good idea
mercutio: there are some situations where ios is a lot better than android
like running a bit of code when a phone has to wake up
acf_: to make desktop applications in html / javascript and ship and entire chrome installation with them
mercutio: so like if your phone is mostly idle... you can turn the cpu on for a fraction of a second and not actually warm up the caches
so you leave your caches cold to use less power
acf_: hm
mercutio: so when you have compiled code it's much easier to stay within a tight cache instruction window
and that can directly transfer to better idle power consumption
although there are other tricks in use too, like wifi will do arps etc with hardware offload
acf_: afaik today Android's Java runtime actually does ahead of time compiling
mercutio: oh it does doesn't it
acf_: although no idea about performance. I'm sure there is still tons of overhead from the java rutime
mercutio: but yeah .. so iphones actually end up having less ram than android phones
for similar performance level
acf_: yeah I'm not surprised
mfw my phone has a higher display resolution and more ram than my laptop
mercutio: heh my tablet is 2560x1600"
my phone is 2560x1440
acf_: my laptop was 1366x768 until earlier this year
mercutio: my monitors are 2560x1440s too
acf_: now 1920x1080
mercutio: my laptop is 1920x1080 too
acf_: phone is 2560xSomething
mercutio: my phone is from 2014 hah
galaxy note 4
acf_: ahh nice
mercutio: it's not a bad phone
i think it has like 3gb of ram
acf_: until last year I had a Nokia windows phone from 2013
if MS had continued supporting the platform I would have continued buying it
mercutio: yeah 3gb
acf_: but oh well
mercutio: my friend who works for google was saying that windows phone was good
acf_: yeah tbh I think people don't give it a chance
because desktop windows sucks, and it has the same name
mercutio: it's odd, he used to (probably still does) use openbsd etc.
so then he comes and tells me windows phone is good
and i'm like hmm..
acf_: haha yeah. I run linux on all my laptops, etc.. and (used to) windows on all my phones
mercutio: apparently it's been canceled
acf_: yeah that was sad
mercutio: so it's good?
acf_: imo it was very good
mercutio: curious
i wonder why it never took off
i suppose lack of apps
acf_: yeah that's probably the big one
mercutio: actually i know someone else who said that it was good but he's a windows fan boy
acf_: my thing is I don't like installing a ton of random apps anyway so I'm fine with it
mercutio: yeh i don't ilke too many apps either
acf_: imo the Android UX used to be awful
mercutio: i hardly use my phone compared to a lot of people it seems
i tried to add a contact the other day and i had to figure out where everything was :)
acf_: like I remember samsung long ago
mercutio: it's odd you go down the bottmo left to contacts.. then to the top right to add
acf_: swiping through the 3 screens of apps trying to find the one I wanted
mercutio: it just seems a weird ui
acf_: on WP, I'd just search the first 3 characters
mercutio: like human eye goes top left to bottom right normally
so a goood way to confuse people is go bottom left to top right
for phones swyping from the left etc isn't actually that bad a way of doing things
pushing your thumb bottom left seems odd though
acf_: also you have a samsung right
mercutio: yeah
so they make everything novel
acf_: iirc samsung replaces all the default android stuff with super shit samsung stuff
mercutio: yeah it does
some of it's nice
acf_: last I checked (galaxy 8) they still do it
and it's still bad
mercutio: but some of it's just odd
acf_: one of my biggest peeves with them
mercutio: yeah i dunno some things are nice on it though
acf_: is their software all updates separately from the both the "system" and the Google store
mercutio: yeah i wish they did more updates
acf_: it's like the least you could do is have a unified update mechanism
mercutio: note 4 has been abandoned pretty much
acf_: all this inconsistency just feels gross
mercutio: and the newer notes are kind of lame
like they took out microsd and replaceable battery
acf_: I heard note 7 is good
mercutio: like i have a replacement battery on my phone
i don't have a microsd, but i like the option none the less.
i used to use a microsd with phones
but i just don't want to store lots on it anyway
acf_: yeah I just get a big internal flash and use that
microsd is kinda slow r/w performance anyway
mercutio: oh wow apparently it does 24bit/192khz audio
yeah there is uhs-1 that speeds things up a bit but it's not supported on most older phones
also sdcard doesn't tend to have good wear levelling
***: r0ni has quit IRC (Read error: Connection reset by peer)
ziyourenxiang has quit IRC (Ping timeout: 250 seconds)
mercutio: brycec: did you notice openbsd 6.5 upgrade was weirdly fast?
i think openbsd must have improved their disk performance...
i can't see anyhting in the release notes about why though
mhoran: It was fast!
I always accidentally pull from the default ftp4 and have to restart from a reasonable mirror
mercutio: oh i did it from mirrors.arpnetworks.com
but the pulling/installing bit seemed to be noticably less sluggish
Hacking on virtio(4), including defines, bug fixing and pci device list.
wonder if that was it
mhoran: Ah nice. Yeah I've got it on a pcengines box
Home router :)
mercutio: ah
is pcengines x86?
mhoran: It's an amd64 clone iirc
mercutio: ah ok
mhoran: Apu.4
mercutio: i saw they just moved mips to clang
oh i think they moved to the clang linker too
so compiling should be faster...
mhoran: Yeah
Every time I upgrade obsd something breaks work dhcpcd, which I need for ipv6 :(
mercutio: odd
mhoran: But the maintainer is great and super responsive
mercutio: they changed openbgpd quite a lot in 6.4
mhoran: I'm probably one of like 3 people using it
mercutio: it's kind of cool that they're active, but yeah can require kind of lot mangling to get things right sometimes
***: ziyourenxiang has joined #arpnetworks
mercutio: has anyone else upgraded to OpenBSD on a VM yet?
mhoran: up_the_irons: something seems wrong with the ARP DNS servers? My monitoring is going crazy because of timeouts and it seems to be to do with reverse DNS resolution.
Changing to 1.1.1.1 resolved.
mercutio: you can tag me too mhoran ;)
reverse dns on what?
mhoran: Postfix.
mercutio: i can resolve 1.1.1.1 fine on both the arp dns servers
mhoran: I had the ARP servers in /etc/resolv.conf and my SMTP checks were timing out. I changed resolv.conf to 1.1.1.1 and they're good again. telnet to matthoran.com 587 was taking ~5s to get a banner, which IIRC is the resolv.conf timeout before it tries another server.
mercutio: weird
yeh the default is stupid high
3 seconds is better now days
***: hive-mind has quit IRC (Ping timeout: 250 seconds)
hive-mind has joined #arpnetworks