ok I would never have expected this
?
I moved a perl bot I am writing that connects to btc-e.com to an arpnetworks vps
from my laptop here at home
instantly it is faster and reacts so quickly it makes me $2.33 USD equivalent in btc on the spot
lol nice
thats just the test account, time to move the big account
hahaha, nice
Soon toddf's VPS will pay for itself
since 12/10 I'm only $3.31 shy of being able to pay my next months bill .. before I move it to arpnetworks, so .. it seems realistic
(my bill is $65/mo fwiw)
toddf well i helped start the #coindev "virtual currency" software / service dev community, and we host with arp
you're both welcome to join our channel, and at the right place! :)
toddf which OS?
openbsd of course
ah, fbsd here
Anyone familiar with how squid operates?
#squid seems quiet right now. I am getting no HITTs in my access log: only TCP_MISS/200
Its set up as a reverse proxy
requests are going through, but not being cached I dont think
A reverse proxy? Any reason not to use nginx/lighttpd/etc?
(varnish..)
brycec is squid truly irrelevant largely, now?
if so, an era has passed
I'd say it's overblown and overcomplicated for most reverse proxying
forward proxying it's still peachy
But Varnish is defintely the defacto caching reverse proxy
nice
brycec: can varnish be setup on say a front end web host with a separate apache server as the backend on a defferent box?
mnathani: Yes
That's what Wikipedia does, for instance
Varnish is awesome
Full disclaimer I've never used Varnish. Looked at it, but purely out of curiosity.
i wish I had the need for it though
ditty
*ditto
url to varnish?
toddf: whoa that's awesome
https://www.varnish-cache.org/
up_the_irons :)
mnathani ty
yea varnish looks useful
pure C or?
I just switched to varnish, but my pages are still not being cached. :-(
boo
I'd say it's time to check that the backend servers aren't sending do-not-cache headers
(tcpdump)
And of course check that varnish is configured correctly :p
Relevant https://www.varnish-cache.org/content/why-isnt-varnish-caching-requests-seem-pass-straight-through
Thanks
Apparently Cookies?
mnathani: easy test on the cookies: cURL
I swear curl is 90% of my web server debugging toolkit
curl ftw
libcurl too, if you're writing programs
how does libcurl help?
er, curl/libcurl
robonerd: rather than writing your own methods to open sockets etc, just use curl
can you get the raw binary somehow?
robonerd: you mean the raw response from the server? yeah of course.
the ppl in #osdev were speaking in the same way about dropbear last night
dropbear is an ssh server, not a client library
well yea
isn't dropbear gone?
i thought i'd heard it was KAPUT
no
i prefer trafficserver to squid or varnish
m0unds: i hope not!
i like for multiple implementations of important projects to stick around
i could have sworn it was dropbear that died
saw that it didn't, now i'm wondering what did die and i confused it with dropbear
dropbear used to have vulnerabilities
and people didn't upgrade versions
and then some people did but didn't upgrade the version number
was there a period where it wasn't maintained well?
or partially abandoned or anything like that?
really ssh servers shouldn't say the version number clearly.
no idea
i use dropbear as a client these days
it works on old hp servers (ilo2) where openssh doesn't
it connects faster over high latency links
ah
and it even connects faster over low latency links
ever tried mosh?
but it doesn't aggregate text as much
so it can be a little slower for some things, and show more constant information
like typing dmesg
in real access, latency seems better if anything
and it's pretty fast
and it's hundreds of msec faster if dropbear on both sides going international
dbclient -l root arp.meh.net.nz pwd  0.00s user 0.00s system 0% cpu 1.963 total
ssh -l root arp.meh.net.nz pwd  0.01s user 0.00s system 0% cpu 2.201 total
that's to openbsd openssh
which is 170msec away
weird ping is usually less than that
it's 1.45 seconds to a linux host at arp
with dropbear server
i still kind of wish the handshaking was quicker
oh mosh was hell slow for me
it was laggy with 10 msec ping
but one of the ends was using atom at the time
and it's very cpu hungry it seems
i thought mosh was a novel idea, but i don't like using hugely inefficient software
i typically only use it on my laptop because connectivity is so variable it saves me from having to reconnect regularly
never really noticed significant cpu utilization at either end
ahh
well atoms are pretty slow
yea
macbook is an i7
it couldn't do samba at gigabit speeds
server i was working with was a dual xeon
ahh, bummer
xeon doesn't mean much :)
it's like saying a gas car :)
anyway, mosh still uses ssh to connect
mid 2012
whatever
so recent
192gb ram
etc
yeah
probably a e5 of sorts
e5-2690 or something maybe
wouldn't have dismissed cpu perf of a 10 year old xeon or something as being a reason for low cpu utilization...
herp derp
oh right
the old pentium4 type xeons were hell slow
err 5000 series
yep, netburst is a joke
bad arch
well about as bad as atom
good at warming up a room though
not really
maybe a small room
3.2GHz prescott was like a 130W cpu
tdp
or something absurd
they still don't use that much power normally
power supplies have got more efficient since then
and there's been some better idle, gating etc support
and cpus run closer to 0% load now
but in reasonable to high usage there's not much diff between old and newer cpus really
like 5000 series to 5400 series isn't huge jump
5400 to 5500 is a bit
5500 to 5600 is tiny i think
then a bit of a jump again from 5600 to whatever the next series was
is that e3 next?
it is isn't it
e3/35
e3/e5
and the affordable e5s aren't much faster than i5-2600s
err i7-2600
and power/efficiency/performance has hardly changed since then again
apparently ivy bridge cpus are more efficient than haswell at load
i suppose amd kind of fell off the train, so intel doesn't have aynone to compete with except arm type things that are coming up
amd still does well in hpc
with video cards?
mostly because of thread count
haven't seen a lot of amd powered gpu compute
nvidia seems to own that space
you mean stuff that powerpc is winning in power/performance?
that's curious
consdiering that amd is much better than nvidia at bitcoin
i assumed that amd was better in general with opencl type things
yea, i think part of it is opencl vs cuda
cuda being more ubiquitous
which is silly
i haven't touched either
it sounded like a nice idea
i've been playing with ssd raid
adn reading up about stuff
i think cuda's been around longer, sorta like physx or whatever in teh gaming world
it seeems intel motherboards can't do more than about 1500mb/sec on onboard sata
so adoption is higher
cos they have a 2GB/sec link for pci-e
intel mobo w/the SATA controller attached at the southbridge or a separate controller like marvell or whatever it was they used?
x4 i suppose
intel onboard direct
i get about 1400mb/sec with x4 pci-e x2
err pci-e 2
i think slightly over 1400mb/sec
so i suppose it basically it is x4 limitations
at x8 i was getting faster speeds
but at those speeds it doesn't amtter too much
but i think i/o bus speeds for non video is likely to be the next significant improvement
video cards are iommu now
err can do iommu
it may not be that far away that there are dual cpus on small computers
with a lower power fanless cpu and a full cpu
that's something AMD's working on, actually
arm + x86
oh, interesting
arm already do that kind of thing
http://www.amd.com/us/aboutamd/newsroom/Pages/presspage2012Oct29.aspx
first step is servers
that's damn cool.
time xterm -e pwd" isn't much faster than it was on a p120.
that + APU stuff (xbox, ps4, low-end laptops) are where they're really kinda pushing stuff a bit
there's lots of inbuilt latency etc limitations
and downclocking big cpus is probably not going to keep latency low
i haven't really considered amd seriously for years
last i knew their memory performance and single threaded performance was way down
yep
multithreaded performance is stellar
and most things don't scale to multicore
yep
which is what i wondered about back with pentium pros
not much scales on linux yet
lbzip2 is damn cool.
compiling scales well
but i think we're really close to needing things like multithreaded file copying
like if you copy 100,000 files
it shouldn't iterate through the files one by one
but on single hard-disks they don't really cope with multiple transactions at once well
but ssd's and raid arrays should be ok
having lots of threads for encoding (single encode process per core) is nice
so that's my personal thing i wnat to look at soon
for multiple files?
yep
yeah multipel files is easy
i don't do any encoding myself
i don't even do mp3 encoding from cd's
i do some audio work for friends, occasionally do video
but cpus are fast enough now that single cpu'ing that is still going to keep up with the cd
yea
i ripped heaps of music years  back
17 seconds to encode a 70 minute album from wav -> flac or wav, flac or whatever -> other format
it's pretty hilarious
heh
doing 8 tracks at once
it used to use up  50% cpu to play a mp3
and i had to fuck around with nice lervels and shit
to make sure it didn't skip
i couldn't play mp3s on the machine i had til i was in HS
cpu was too slow, 486DX-33
haha
haha
i went 486dlc40 to p75 i think
err
it was 486dlc40 to some amd thingy
like a fast 486
with a weird name
then to a p120
i went 486dx-33 4MB RAM -> 20MB RAM -> 300MHz AMD K6-2
i overclocked the p120 to 133 mhz and removed the fan
so much headroom in those days
4mb ram gotta suck
it was cutting edge in 1993
double speed cdrom too
heh
shit 91
mp3s were around in 91?
oh you said 93
how did i read it as 91
yeah, i had that machine til 1999
hahaha
heh
i got amd k6-2 sometime too
with 128mb of ram
cdrom interface was via the isa soundcard
i had 24mb before that
and couldn't run X
i didn't have enough vram in the 486 to run x at higher than 640x480
heh
did you overclock isa bus?
nah
i didn't have enough memory
i think i had s3 virge
i just used svgatext mode
my 486 was 1MB cirrus logic or something
i could get 800x600 in wfw 3.11
and had like 100 columns or so
heh
i dunno how much memory i had
probably 2mb?
i think it was s3 trio64
i have a pci version of one of those somewhere
s3 trio or verge maybe
lots of people had vergs
trios were cheaper
i went to radeon 9000
which could do 1920x1440?
but on crt that flickered heaps
i usually just used 1280x1024
yeah, probably low refresh or whatever
yeah, same
and then got annoyed with the lack of space sometimes and jumped up to 1600x1200
yeah 1280x1024 was 85 hertz
and 1920x1440 was like 60 hertz
1600x1200 75 hertz or something
19" crt
but when youwent up resolution things got less crisp
but yeah everything uses up so much memory niow
  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND
17037 root      20   0  333392 100136  80780 S   2.7  1.2   0:09.62 Xorg
i have 5 xterms open, and pidgin
but i was surprised how well linux seems to deal with 8gb of ram
compared to windopws
sometime memopry volumes went up enough that it doesn't really matter now
i've got 2GB free in win7 right now on this particular workstation
on 8gb ram?
yep
i have 5 gb free on 16gb right now
on the windows box
but i'm running chrome
probably superfetch caching stuff
and skype and putty
i disabled that stuff
why?
cos samsung told me to
lol
samsung magician software suggests you disable it
hahah
my corsair SSD has 3 years of use and still indicates 100% wear level with like 8TB of writes
i dunno i had 8gb for a while in it
cos i was curious how it'd go
(i default to thinking 16gb makse sense thse days)
i wonder if it's a write-saving measure for samsung
yea, my home ws is 16gb
this one's at my office
and i noticed the upgrade to 16
it's cos i junmped from 6 to 16 and thought 8 may have been enough
but i'm in similar situation with my linux bvox too
i have ram lying around
i could upgrade to 16gb or 24gb on server
but it's on 8gb
and i run chrome on it too
and it deals much better on 8gb
and i dual boot the windows box with linux too
and i notice no diff between 8/16
my original justification was for virtualisation though
and i was doing linux and windows on the same box for a bit
and upgrading from 16 to 24gb helped
samsung suggests disabling superfetch w/ssds as a memory saving measure
oh interesting
i was curious why they'd suggest it - iguess it makes sense in lower memory ssd envs like laptops or whatever
so i should reenable it
like if you've got a notebook with 4gb of ram, it might make a diff
the biggest difference i find with more ram is uncompressing large files yhou've juist copied on
but even then, i dunno
like say you have a 4gb archive
that you just copy on
it's nice if that 4gb stays in memory
but generally speaking it's still going to be slow