#arpnetworks 2013-05-17,Fri

Hein: waiting is kindof par for the course here, but whatever the wait, it has a good reason, and once your service is up, well, you will enjoy it ;-)

[08:28]

hazardous

toddf: i enjoyed mine :>
this is the first host where i haven't had to contact support pretty much ever
not even during setup after i got my vpn certs

[08:29]

toddf

automation is the name of the game, kindof mandatory for a not 24/7 staffed facility .. and indeed it is awesome to be entirely self service
when I first was testing my first vps, I seriously would debug and crash it because it was easier to reboot than my physical system across the room, it dump'ed to swap faster, and rebooted faster, and was .. faster .. (compiled faster, disk io better, md5 -t faster) than any physical system I own even to this day ;-)

[08:30]

hazardous

toddf: tbh i wish i could get second dedi @ arp

[08:31]

toddf

read 'debug and crash' -> testing questionable bleeding edge kernel diffs

[08:31]

hazardous

but my hardware needs have scaled up significantly
and i can't afford here :<

[08:31]

toddf

you can't afford an arp metal system or an arp vps system? thats surprising

[08:32]

hazardous

metal
i'm eating around 103gb ram atm

[08:32]

toddf

I have not found many offerings that compare price wise

[08:32]

hazardous

last week spiked to 4684 simultaneous online users
lol

[08:32]

toddf

when you're eating that much mem you kindof need to supply the hardware your self

[08:33]

hazardous

yeah, figures
i'm nowhere near LA unfortunately
and i don't have a capex budget at all
(this is a personal hobby site i run for fun and originally made in a few hours on a weekend)

[08:33]

toddf

you can always purchase and ship to LA, I bet up_the_irons or helpers might charge a reasonable fee for assembly
google ads can produce some revenue if you're looking to fund your obviously interesting hobby

[08:33]

brycec

Plus the cost of colo, which would probably be a bit higher than metal.

[08:34]

hazardous

i'm running one ad above the fold and one ad at the footer
it's producing enough to cover the single arp metal dedi i have

[08:34]

RandalSchwartz

some people must have odd hobbies to need so much ram

[08:34]

brycec

nice

[08:34]

hazardous

though flash ads piss me off immensely

[08:34]

RandalSchwartz

RandalSchwartz is currently contracting at an ad syndication shop

[08:35]

hazardous

and adsense doesn't seem to have a function that serve 'media ads like gif are fine, but not flash'

[08:35]

brycec

heh was about to say something like that. It seems weird to me that 5000 users would need 100+GB of RAM

[08:35]

hazardous

brycec: another thing you'll find amusing is that i do more in than out bw-wise

[08:35]

RandalSchwartz

dealing with 1.2 billion hits a day coming in to 130 servers with a SLA of under 0.2 seconds to compute qualified ads
it's actually pretty interesting. downtime is expensive. :)

[08:35]

hazardous

RandalSchwartz: my friend did something like that, i think they used erlang or something
and not a 'traditional web language'

[08:36]

RandalSchwartz

this is all mod_perl

[08:36]

brycec

that is interesting

[08:36]

RandalSchwartz

mod_perl talking to couchbase

[08:37]

hazardous

this is titillating

[08:37]

RandalSchwartz

oh - and those 140 servers are all 24 core
130

[08:37]

brycec

RandalSchwartz: geographically spread out?
(the servers, not the cores :p)

[08:37]

RandalSchwartz

for some reason, the first one is numbered 010... so I keep think "up to 140 is 140", but not
no - all in one data center
not much CDN yet

[08:37]

hazardous

i think maybe i'd use less ram if i wasn't using php

[08:38]

RandalSchwartz

php - there's your problem. :)

[08:38]

hazardous

or scala

[08:38]

brycec

ouch, yeah... you could go much leaner

[08:38]

RandalSchwartz

php is a perl wannabe :)

[08:38]

hazardous

maybe drop apache for something else too

[08:38]

RandalSchwartz

got a lot of things wrong that perl got righ

[08:38]

brycec

Anyone else see they ported Perl to Javascript? jsfuck.com

[08:39]

hazardous

also it's not really an interesting hobby, just some small league of legends fansite
brycec: didn't they port unreal engine to js

[08:39]

RandalSchwartz

actually - there's a perl6 to javascript compiler in the works

[08:39]

hazardous

via emscripten or something
llvm -> js

[08:39]

RandalSchwartz

I'm looking at developing things in Dart in the near future

[08:39]

brycec

Apache is definitely a hog by default... I prefer lighttpd, but maybe Apache works better at that scale

[08:39]

RandalSchwartz

especially after my interview with one of the Dart guys for FLOSS Weekly

[08:40]

hazardous

yeah, they did port unreal engine 3 to js

[08:40]

brycec

nice

[08:40]

hazardous

that sounds like a horror in itself
decent fps in firefox, crap fps in chrom{e,ium} which is strange, usually the opposite

[08:40]

RandalSchwartz

I've got one lightweight site using nginx directly, without apache

[08:46]

***

betojsp has quit IRC (Quit: Page closed)

[08:46]

brycec

All of my sites are pretty lightweight, essentially I'm the only visitor, mostly just web/api backends... A great application for lighttpd. I'm sure nginx could do well too, but I see them as about equals, and both fantastically lightweight which is what I care about.

[08:51]

And since GAE now supports PHP, I'm thinking of moving a lot of that over to GAE... just f or kicks. And redundancy.

[08:56]

hazardous

yeah i have a lot of small stuff that could work fine
i'm actually trialing cherokee atm
for my web frontend (which is just a giant template filled out by json)

[08:58]

................ (idle for 1h18mn)

GluffiS

nginx is nice... coupled with varnish you get really nice speed on static content

[10:16]

........ (idle for 35mn)

***

scottschecter has quit IRC (Quit: WeeChat 0.4.0)
scottschecter has joined #arpnetworks

[10:51]

....... (idle for 30mn)

up_the_irons

hazardous: wow, that's a lot of ram

[11:22]

hazardous

up_the_irons: there are some really really cheap 64gb/128gb places, for <$200
one of the providers is on nearly pure he.net bandwidth
but somehow has no native ipv6
(wtf)
also i presume you saw my ticket reply about the seagate?
it's not particularly important, i have a full backup and already dropped it out of my roundrobin so basically "anytime" is fine

[11:23]

up_the_irons

hazardous: yeah i'm going to be replacing that tonight and hopefully i'll also have the SSDs for Hien

[11:24]

hazardous

btw, the existing disk (the hitachi), smartctl is complaining about and i'm not sure why

[11:24]

up_the_irons

hazardous: cheap 64/128GB for less than $200? wow

[11:24]

hazardous

up_the_irons: yeah, look on wht for webnx's offers
they don't permit irc, refuse to swip, don't have native ipv6

[11:25]

up_the_irons

jeez

[11:25]

hazardous

they have an unmetered bw subsidiary, gorillaservers
it's almost entirely he.net bw
and they somehow don't have ipv6

[11:25]

up_the_irons

dumb

[11:25]

hazardous

which is incredibly confusing and hilarious
do you want to know the best part
By enabling IPMI I accept all responsibility & liability for managing and securing the IPMI function and indemnify GorillaServers Inc. for any damages that may result from enabling IPMI on my server(s). GorillaServers Inc. does not guarantee stability or function of the IPMI service, and replacement of motherboards for IPMI purposes is not guaranteed. IPMI is a bonus feature not covered by SLA.

[11:26]

GluffiS

I wonder when common sense died...

[11:27]

hazardous

dunno, probably when openvz was released :>

[11:28]

GluffiS

haha
i really dont like openvz...

[11:28]

hazardous

same
except the term would be stronger than dont like

[11:29]

GluffiS

:D solaris containers are way better imho

[11:30]

hazardous

i forget the name
zones?

[11:31]

GluffiS

yes

[11:32]

RandalSchwartz

those were designed based on freebsd jails, as i recall.

[11:39]

up_the_irons

hazardous: lol on IPMI

[11:40]

hazardous

up_the_irons: 'lol' was pretty much my thought too, followed by 'wait, they're serious?' a few minutes later

[11:41]

GluffiS

RandalSchwartz: probably. Don't know the story behind zones. they worked great the last time i used them a couple of years ago

[11:41]

up_the_irons

lol

[11:41]

staticsafe

LXC is worth looking at

[11:42]

hazardous

oh ya
here was the thing

this is my existing hd (hitachi), http://pastie.org/pastes/7921879/text?key=lj2ji5mvnhslwhuvkcei9a

smart kept whinging about prefail, dunno what that means exactly

[11:42]

GluffiS

staticsafe: well, I kinda left the IT business a year ago ;)

[11:43]

hazardous

anyhting to be worried about or does that look ok

[11:43]

up_the_irons

hazardous: i'm honestly not good at reading smart status. when i look for hd pre-failure, i see how many reallocated sectors there are, which is reported by the raid cards we use

[11:47]

hazardous

ah, k

[11:47]

.............. (idle for 1h9mn)

toddf

I bet solaris zones are based on the hardware zones in sparc T* series systems that can literally continue running the various zones even when the master reboots

[12:56]

GluffiS

toddf: there is a difference between zones and containers... container are like chroot/openvz, and zones survive master reboot ;)

[12:59]

toddf

gluffis: zones on an x86 system?

[13:01]

GluffiS

no
zones is only on the T series i think

[13:01]

hazardous

CVE-2013-2094 when run inside an openvz 'vps' takes out the entire host

[13:01]

......... (idle for 42mn)

up_the_irons

lol

[13:43]

..... (idle for 20mn)

brycec

<3 that one

[14:03]

....... (idle for 34mn)

***

staticsafe has quit IRC (Ping timeout: 264 seconds)

[14:37]

staticsafe has joined #arpnetworks

[14:46]

.................. (idle for 1h25mn)

Hien

up_the_irons: hello, you here?

[16:11]

..... (idle for 23mn)

mercutio

does anyone know any good sites explaining ipv6 addressing realyl clearly / simply
you know, things, like allocating a /64 for each network
operational stuff rather than saying how many ip addresses people will get

[16:34]

hmm i found an apnic presentation that suggests each site should have a /48

[16:41]

staticsafe

sites have /48s, subnetworks within that site get /64s

[16:42]

mercutio

and i think it was suggesting that mobile phones should have a /64
maybe for tethering
so what's considered a site? is a dsl connection a site?
are people normally doing /48s for dsl, or /56s, or /64s ?
scanning will be fun :/
actually maybe it's better as far as scanning goes
makes it harder for people to scan

[16:42]

.......... (idle for 46mn)

***

cpet has joined #arpnetworks

[17:29]

cpet

up_the_irons: ping

[17:30]

***

cpet has left

[17:41]

........... (idle for 54mn)