***: gcw|mini1 has joined #arpnetworks
LT has joined #arpnetworks
gcw|mbpro has quit IRC (Ping timeout: 246 seconds)
Webhostbudd has quit IRC (Ping timeout: 250 seconds)
fink has joined #arpnetworks up_the_irons: jlgaddis: no web frontend to powerdns, it just powers my Reverse DNS manager
jdoe: mercutio : the comparable Intel stuff is sexy too, but way expensive mercutio: up_the_irons: do the new cpus support bulldozer?
err i mean are they bulldozer
i actually have no ideahow much cpu ssh would be taking up
but i like the idea of it being accelerated :0 up_the_irons: mercutio: they are bulldozer, yes. would ssh even take advantage of the hardware aes-ni? like, how would it know it's there.. probably need a new version of ssh LT: if it's anything like the old VIA AES instructions it just requires the right openssl setup for most things to use it up_the_irons: ah cool LT: 'openssl engine' ***: irons|xchat has quit IRC (Ping timeout: 246 seconds)
HighJinx has quit IRC (*.net *.split)
arenlor has quit IRC (*.net *.split)
teneightypea_ has quit IRC (*.net *.split)
dferris has quit IRC (*.net *.split)
Igneous has quit IRC (*.net *.split) mercutio: up_the_irons: openssl needs to be up to date ***: arenlor has joined #arpnetworks mercutio: lt: ubuntu doesn't show it in openssl engine now
do you have one of those cpus now? ***: HighJinx has joined #arpnetworks
dferris has joined #arpnetworks
Igneous has joined #arpnetworks mercutio: openssl speed -evp aes-256-cbc
you want to run that command
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes ***: teneightypea has joined #arpnetworks mercutio: aes-256-cbc 202878.63k 208678.91k 206311.52k 225509.62k 222816.90k
if it's anything like as high as that it's accelerated
if it's more like:
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-256-cbc 73896.75k 74157.50k 75189.17k 75321.54k 75582.84k
then it's not LT: hmm rhel stilll shows it in engine
aes-256-cbc 614184.36k 684397.85k 696434.09k 699525.12k 701133.83k mercutio: that's accelerated numbers :) ***: fink has quit IRC (Quit: fink) mercutio: actually i wonder if that first one is accelerating properly
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-256-cbc 164786.10k 183756.50k 189753.15k 190754.71k 191059.91k
cos that's the number of non accelerated core 2 duo LT: actually mine wasn't accelerated... mercutio: really?
how come?
options:bn(64,64) rc4(16x,int) des(idx,cisc,16,int) aes(partial) blowfish(idx)
# openssl engine
(rsax) RSAX engine support
(dynamic) Dynamic engine loading support
is there something else that has to be done too? LT: hmm... you have me wondering now mercutio: the first number is xeon e5420
actually xeon E5345
model name : Intel(R) Xeon(R) CPU E5345 @ 2.33GHz
and you''re like nearly 10 as fast
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-256-cbc 134099.06k 152144.10k 157358.49k 158168.27k 157409.28k
e5420 is quite a bit faster
actually i wonder if i'm accelerated properly
https://bugs.launchpad.net/ubuntu/+source/openssl/+bug/1001424
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-256-cbc 213802.95k 687449.25k 1094622.32k 1083138.24k 1088128.52k LT: oh yeah it is accelerating, I thought I could trick it by setting -engine to something else but it seems to load the engine anyway mercutio: it's a lot quicker when you pass -decrypt
-decrypt time decryption instead of encryption (only EVP).
The 'numbers' are in 1000s of bytes per second processed.
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-256-cbc 59636.47k 41641.90k 141532.70k 150321.79k 152577.39k
but e5345 still sucks at decrypt LT: aes-256-cbc 613619.75k 1452257.90k 1860791.98k 1988995.41k 2028642.30k mercutio: lt: what cpu is that? LT: X5690 mercutio: hmm
oh that's really high ghz isn't it?
3.47ghz ?
versus 2 ghz
it's the first intel series to support aes-ni i think
wow those things are expensive :/ LT: yes... but useful if you're stuck running something single threaded mercutio: like dos?
the new e5s latency reduction thing sounds intersting
i suppose it'll come into desktops soon
opteron 6276 seems similar speed to x5650
e5-2690 is like 50% faster
but e5-2690 is like what the recent version of what you're using
ie: expensive :/
it's also power hungry
http://www.anandtech.com/print/5553
reading that
what opteron 6276 is expensive too :/ ***: fink has joined #arpnetworks
fink has quit IRC (Client Quit) mercutio: oh it's cheaper than X5670
and X5690 isn't even sold here :/ LT: the latest new thing always is expensive - we tend to wait until the next new thing is out then buy up the best of the last generation on the cheap mercutio: ahh i see
i prefer lower spec newer stuff generally
but end up with lower spec older stuff :/
the e5-2620 is actually pretty good value
and all the opterons seem to be branded hp
up_the_irons: what cpus are in the kvm nodes atm? kvm hides the cpu type
it says 2.641 ghz
but that's about as much
x3450?
i'm guessing it's 2.66ghz
or e5430
or e5150
you have to tell ssh to use aes
otherwise it defaults to blowfish
without acceleration aes256-cbc and blowfish are about the same speed and aes128-cbc is faster
and ptuty defaults to aes rater than blowfish LT: openssh defaults to aes128... mercutio: does it?
aes128-cbc?
Ciphers aes128-ctr,aes192-ctr,aes256-ctr,arcfour256,arcfour128,aes128-cbc,3des-cbc
i actually foudn this
commented out
blowfish isn't even in that last?
but if i change that first entry to be aes128-cbc ssh to localhost with dd'ed zero data is faster
oh hmm
that really isn't a good test is it LT: default is in the ssh_config man page... probably varies due to version, but aes has been preferred for many many versions mercutio: it's aes128-ctr versus aes128-cbc?
what's the diff
i was reading something old that said blowfish default :) LT: one is ctr mode one is cbc mode :p
ctr is meant to be easily parallelisable, cbc isn't mercutio: 209715200 bytes (210 MB) copied, 6.54909 s, 32.0 MB/s
versus
209715200 bytes (210 MB) copied, 2.4191 s, 86.7 MB/s
and non-zero file
oh?
first is ctr second is cbc
dd if=/var/www/testfile.zip bs=1024k | ssh -c aes128-cbc root@localhost 'dd of=/dev/null'
hmm# gzip -cv /var/www/testfile.zip| wc
/var/www/testfile.zip: -0.0% 820474 4625767 209749312
so non compressable
it would be nice to be like more than 50% faster than that :)
so that ssh can be liek gigabit speeds with encrpytion LT: if you want fast ssh you need the hpn version - that actually has a parallel version of aes-ctr mercutio: oh?
i don't really need faster ssh
i was just curious how much diff these acceleration things made
and i'm not getting a lot of diff
aes128-cbc is faster than aes128-ctr on xeon e5435 too though.. LT: maybe there's no acceleration for ctr but there is for cbc? dunno mercutio: does the hpn version work on openbsd?
xeon 5435 has no acceleration aynway
err i meant 5345
but i think that could be right too LT
cos i foudn a rhel bug about such, i cant' acecss it though to see more about it LT: cbc is a pretty standard way of doing things that's been around for ages, but ctr is newer and I think the ssh people might even have invented it mercutio: hmm
well the ssh people took out the none cypher
i like the idea of encrypting auth but not data :) LT: the hpn version adds none back mercutio: ahh
i do slightly remember hpn from ages back
when ssh had window size limitations
that made international transfer speeds always suck
back then it was always behind ssh by quite a few verisons LT: http://www.cl.cam.ac.uk/~rja14/book.html is a great free book for understanding the differences between cbc, ctr, etc as an aside mercutio: cool i'll bookmark it
hmm dd over ssh is faster than scp ***: ix33 has quit IRC (Read error: Operation timed out) mercutio: i do actualyl want to understand encrpytion sometime .. especially with regards to interactive ssh type applications ***: notion has quit IRC (Read error: Operation timed out) mercutio: (ie not much data.. data could be guessed by delays between keystrokes etc) ***: notion has joined #arpnetworks
ix33 has joined #arpnetworks
LT has quit IRC (Quit: Leaving)
hive-min1 has quit IRC (Ping timeout: 268 seconds)
hive-mind has joined #arpnetworks up_the_irons: type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-256-cbc 279385.84k 277993.37k 301771.51k 318107.50k 319316.99k
mercutio: ^^
AMD 6216
mercutio: the kvm nodes are dual E5430's
these are numbers on the E5430's (no acceleration)
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-256-cbc 95563.29k 106319.22k 110646.64k 112421.96k 106064.11k ***: fink has joined #arpnetworks
HighJinx has quit IRC (Quit: Computer has gone to sleep.) up_the_irons: jeez
Intel Xeon E5-2660 Sandy Bridge-EP 2.2GHz ***: irons|xchat has joined #arpnetworks up_the_irons: $1329
all i can is
Fuck That(tm)
*say is ***: hive-mind has quit IRC (Ping timeout: 272 seconds)
hive-mind has joined #arpnetworks
hive-mind has quit IRC (Ping timeout: 240 seconds)
hive-mind has joined #arpnetworks
fink has quit IRC (Quit: fink)
hive-mind has quit IRC (Ping timeout: 268 seconds)
HighJinx has joined #arpnetworks
atmark has joined #arpnetworks
fink has joined #arpnetworks
hive-mind has joined #arpnetworks
hive-min1 has joined #arpnetworks
hive-mind has quit IRC (Ping timeout: 245 seconds)
fink has quit IRC (Quit: fink)
fink has joined #arpnetworks
gcw|mini1 has quit IRC (Remote host closed the connection)
fink has quit IRC (Client Quit)
gcw|mbpro has joined #arpnetworks
Webhostbudd has joined #arpnetworks
fink has joined #arpnetworks
mtve has quit IRC (Ping timeout: 244 seconds) phlux: 3:28PM up 391 days, 17:36, 4 users, load averages: 0.05, 0.06, 0.01 ***: gcw|mbpro has quit IRC (Ping timeout: 272 seconds) phlux: ^My ARPNetworks FreeBSD VPS...I think I'm going to be starting it from scratch soon :( up_the_irons: $ uptime
13:49:07 up 958 days, 13:37, 3 users, load average: 4.09, 4.76, 4.38
kvr04 went up to 983 days, my longest box ever phlux: good lord up_the_irons: maybe this one will surpass it phlux: am I on that box, by chance? up_the_irons: that's kvr05
check your vnc host, says what box you're on phlux: I honestly don't even know what I'm on
Roger that, one second
bah
I don't know my VNC host :[13:33] <phlux> I thought it would be in the dashboard somewhere
oh wait, it is
I'm on kvr06 up_the_irons: ah -: phlux is kevinthompson up_the_irons: that's a long running host too
$ uptime
13:07:43 up 934 days, 10:01, 2 users, load average: 3.10, 3.23, 3.31 phlux: nice
Hopefully it keeps it up :P up_the_irons: :) twobitha1ker: all that time with no kernel security patches :D phlux: My ARPNetworks VPS is the hub of my entire network because of its stability
I've got VPSes from elsewhere, and they don't really compare imo
I've considered moving my entire network to ARPNetworks, but I don't know how much sense that would make for me. up_the_irons: phlux: :)
phlux: i say "do it" phlux: haha, I bet!
If I could ensure that my servers were all on different hosts, I will be convinced to swap over everything. up_the_irons: twobitha1ker: i knew someone would say that... phlux: Someone says that to me about my VPS, too
I tell them to go ahead and exploit it for me right quick up_the_irons: that's what you always hear when u try to tout uptime
phlux: LOL hahaha phlux: They always answer "Sure give me a shell." It doesn't work that way. If I did that, I'd be giving you a head start. up_the_irons: phlux: well, we *do* put VMs of the same customer on different hosts, by default. phlux: No kidding?
So if I picked up 3 more at the same time, all 3 would be on separate hosts? up_the_irons: this one dude got like 15 VMs, in one day, and that was the only time it was hard to make sure they all got spread out
phlux: sure of course phlux: If so, I am going to make that change on Sep 1 (which also happens to be my birthday, yay me!)
Awesome
It's going to be so much easier paying one company :P up_the_irons: phlux: i understand the concern about not having all eggs in one basket, i practice that myself, so why not do it for the customers? fink: phlux: see if you can negotiate for an arpnetworks tshirt -: fink has tried yet failed :( phlux: up_the_irons: I'm glad to hear that you do it that way.
lol fink
YEAH
3 servers for 1 T-shirt
whaddya say?! up_the_irons: phlux: i wish i knew that was your only concern, i would have addressed it earlier :) phlux: up_the_irons: No worries. I originally signed up a long time ago with the intention of "testing the waters," and I ended up using that VPS for *everything* up_the_irons: phlux: haha cool phlux: Speaking of which, I'm teaching a guy some things, and he's most likely going to pick up a VPS from ARP in the coming days.
Hoping his introduction to programming goes smoothly..he wants to be an IT Commisioned Officer in the USCG. I'm waiting for him so we can try to go to OCS together. We're both already enlisted, so we wouldn't really be competing with eachother. up_the_irons: I got invited to OCS, then saw the age cutoff was 29. I officially felt old (I'm 33)
i hunger, cd $food phlux: Dang I didn't know that the cutoff was 29
I'm only 23 (as stated earlier, 24 soon) so I haven't had to worry about age limits for anything, yet kraigu: up_the_irons: child :D
I was considering re-careering here as a direct entry officer, but the Canadian Forces dropped signing bonuses and I couldn't afford the pay cut without one phlux: Man, if I'd have joined a year earlier, I'd have got a signing bonus with the CG.
I became an MST (Marine Science Technician)
I deal with oil spills, HAZMAT releases, waterfront facility regulations, foreign and domestic vessel inspections, and container inspections.
That's a mouthfull. kraigu: also I suggested this possibility to my wife and her reaction was nearly as violent as I could expect in Kandahar :[13:50] <milki> o.O
? ***: hive-min1 has quit IRC (Ping timeout: 240 seconds) arenlor: up_the_irons: Given that kvr05 has 934 days uptime, how is its health? ***: hive-mind has joined #arpnetworks up_the_irons: kraigu: lol
arenlor: health looks good; there are a couple drives with reallocated sectors, but i did a raid verify (twice) on it, and everything checked out; so, i'm just watching it
phlux: that sounds interesting, actually arenlor: up_the_irons: Alright, given that my important eggs are on it, don't care to have it go down too suddenly ^_^ up_the_irons: phlux: you will have to worry about age limit for American Idol soon (26, i think ;)
arenlor: i hear ya. i don't want a repeat of kvr04, where i spent all day and night at the data center nursing it back to health ***: Webhostbudd has quit IRC (Quit: Leaving) ryk: up_the_irons: well they do say natural breastfeeding is better than the bottle, so good job up_the_irons: lol ***: mtve has joined #arpnetworks
fink has quit IRC (Quit: fink)
dj_goku has joined #arpnetworks
fink has joined #arpnetworks
atmark has quit IRC (Ping timeout: 252 seconds)
fink has quit IRC (Quit: fink)
Webhostbudd has joined #arpnetworks
hive-min1 has joined #arpnetworks
hive-mind has quit IRC (Ping timeout: 244 seconds) mercutio: up_the_irons: e5-2620 is a lot cheaper than e5-2660
and your e5430 results seem a little low
ie i got higher results than that from e5420
it may be openssl version, or it may be load / context switching overhead ***: DataCable has joined #arpnetworks
dj_goku has quit IRC (Ping timeout: 244 seconds)
DataCable has quit IRC (Client Quit)
CaZe has quit IRC ()
dj_goku has joined #arpnetworks
dj_goku has quit IRC (Changing host)
dj_goku has joined #arpnetworks
dj_goku has quit IRC (Ping timeout: 244 seconds)
HighJinx has quit IRC (Quit: Computer has gone to sleep.)
Webhostbudd_ has joined #arpnetworks
Webhostbudd has quit IRC (Ping timeout: 250 seconds) up_the_irons: mercutio: well, that was on a production server, hardly idle mercutio: yeh load could create that
it could be related to openssl version too
OpenSSL 1.0.1c 10 May 2012
is what i was using
err actually it was 1.01
1.0.1
i think openssl changed a bit recently ***: fink has joined #arpnetworks
hive-mind has joined #arpnetworks
hive-min1 has quit IRC (Ping timeout: 260 seconds)
hive-mind has quit IRC (Ping timeout: 240 seconds)
hive-mind has joined #arpnetworks
hive-mind has quit IRC (Ping timeout: 245 seconds)
hive-mind has joined #arpnetworks up_the_irons: mercutio: i was using 0.9.8something jpalmer: up_the_irons: less talking.. more testing centos images or spending time with your family.
;) -: jpalmer ducks and runs up_the_irons: i'm researching ways of exporting a block device from one host to another (say an lvm volume from one host to some /dev/foo on another). i've used AoE before, but things these days seem to be leaning toward iSCSI. Anyone have any thoughts on the matter?
jpalmer: lol
jpalmer: the kids are asleep:)
same with the wife jpalmer: iscsi is the new hotness. do it! up_the_irons: brb jpalmer: I think I'ma head to bed.
night all up_the_irons: 'nite mercutio: up_the_irons: think newer versions have improved by a bit
up_the_irons: with iscsi you want to have seperate network for it and use large mtus
balancing can be annoying, you c an use either multiple gigabit or expensive 10 gigabit
but with gigabit you limit top speed to 110mb/sec or something
fibre channel is the other option that is affordable
but most of the affordable stuff ise second hadn
and 10 gigabit ethernet is coming down in cost
that said 100mb/sec is racing, if access times are good up_the_irons: mercutio: i c
mercutio: separate network is no problem, large mtus no problem
pretty easy to bond 2 gige's, but fc would be better
mercutio: have you used any iscsi targets that you would deem reliable? mercutio: only opensolaris up_the_irons: interesting mercutio: and that's in state of "uhh what's happening" up_the_irons: wut?:) mercutio: well when oracle bought sun
everyone left
err a lot of the higher up people left
there was a fork to illumos up_the_irons: ah Webhostbudd_: the illumos camp is hopping mercutio: openindiana is going between prereleases... Webhostbudd_: it's the only place to be up_the_irons: oh, would you look at that... who has illumosvps.com (me) mercutio: openindiana seems stable with ok hardware
stable as in reliable Webhostbudd_: they have a lot of sun people working there so we should see some good stuff mercutio: but not stable as in "one version for years"
yeh Webhostbudd_: well, no mercutio: smartos sounds interesting too Webhostbudd_: smartos is a great concept up_the_irons: dang, i have openindianavps.com too mercutio: md3000i is around for iscsi support but is slow up_the_irons: i musta been looking toward the future... ;) mercutio: there are higher cost options too
hmm
well joyent ported kvm to illumos Webhostbudd_: it runs very well up_the_irons: holy shit, they did? Webhostbudd_: yes mercutio: yeh up_the_irons: wow
runs good? Webhostbudd_: that's what i have heard mercutio: the base tools haven't had the work they need
but the raw support is stable Webhostbudd_: yes
exactly
once they port libvirt mercutio: even though they ported it quickly Webhostbudd_: it is going to be very attractive mercutio: showing that they have good engineers Webhostbudd_: they have a lot of the old sun guys
because of the mass exodus from oracle mercutio: by good poeple leaving oracle
people like the guy who made java left Webhostbudd_: mhmm up_the_irons: smartos _does_ sound interesting Webhostbudd_: all of the zfs, dtrace and other solaris kernel specialists left mercutio: and said how he was unhappy with the direction oracle were taking sun
and you'd think someone like that isn't likely to be someone who would motuh off randomly Webhostbudd_: they aren't
they LEFT mercutio: http://www.eweek.com/c/a/Application-Development/Java-Creator-James-Gosling-Why-I-Quit-Oracle-813517/
yeh but i mean it's not like a straight business man leaving or something
and saying about his new venture Webhostbudd_: no, they just don't want to be a part of the lawnmower
up_the_irons: i could totally see you running smartos one day
up_the_irons: assuming it becomes stable up_the_irons: Webhostbudd_: i'm checking it out right now Webhostbudd_: what's funny is that their port of KVM requires the use of zones
so each kvm instance is wrapped in a zone
oddly enough it prevents a lot of kvm exploits
because the attacker can't break the zone mercutio: i think that was a sensible move Webhostbudd_: yes it was
i mean, why not
it gives the administrator even more control over the virtual instance mercutio: and with zfs you can clone file systems
and do copy on write Webhostbudd_: zfs is amazing mercutio: so it's not as high overhead having heaps of zones Webhostbudd_: smartos is a killer combination of technologies
i would love to see it get bigger mercutio: webhost: marketing on the other hand...
have you seen their videos? Webhostbudd_: no? mercutio: they seem tacky Webhostbudd_: it's pretty hard to market it at this point though
let's be honest, they don't have a product to ship mercutio: i think they do?
joyent are doing private clouds already up_the_irons: so does smartos run illumos underneath? mercutio: http://joyent.com/solutions/privatecloud/ Webhostbudd_: no one who is concerned with how they market a product is going to call that a product mercutio: up_the_irons: yeh Webhostbudd_: up_the_irons: yes mercutio: and is one of the major contributors up_the_irons: and illumos is an opensolaris fork? Webhostbudd_: illumos is the kernel if i am not mistaken
or do they do userland work? mercutio: illumus is like the base system
like netbsd etc i think Webhostbudd_: yea, that was always my impression up_the_irons: yeah i think i get it Webhostbudd_: except they don't actually assemble any of it mercutio: openindiana is more like opensolaris Webhostbudd_: i wanna say it would be like GNU finishing a kernel mercutio: there's also nexenta
or something
which uses apt-get Webhostbudd_: the way i see it, GNU = illumos and OI / nexenta are distros
IF GNU had a kernel up_the_irons: so, with zones, i've always thought of them to be the solaris equivalent of freebsd jails. and since they are just processes that share the same base OS kernel, is it true the only thing you can put in a zone is another illumos instance? Webhostbudd_: up_the_irons: mostly true mercutio: up_the_irons: i think it's like freebsd jails, but a bit more brokenout up_the_irons: (and that instance could run kvm, yeah, but in the end, it's still illumos) Webhostbudd_: yes up_the_irons: roger Webhostbudd_: but there is probably work to emulate linux / bsd systemcalls
i know freebsd jails can run centos
with linux emulation mercutio: web: i think that work is still behind Webhostbudd_: maybe mercutio: smartos is getting a new package manager i think Webhostbudd_: i don't use zones so i wouldn't know mercutio: and packages can be old at present Webhostbudd_: well, all of the illumos stuff seems to be using pkgsrc up_the_irons: so illumos is only the kernel, and openindiana is kinda like kernel + userland (e.g. GNU) mercutio: but i think in general it's under quite a lot of active work up_the_irons: freebsd jails can run cent? LOL Webhostbudd_: up_the_irons: yea, i think that is probably more correct
up_the_irons: bottom line, you don't run illumos up_the_irons: Webhostbudd_: yeah makes sense, at least roughly mercutio: openindiana has both sun and gnu userland Webhostbudd_: directly
yea mercutio: ls you have /usr/gnu/bin/dd
and /usr/bin/dd
and the normal dd doesn't support things like conv=fdatasync Webhostbudd_: there is a really cool fork out there
which im really digging
stormos
debian userland + illumos kernel mercutio: webhost: what have they done? Webhostbudd_: similar to debian kfreebsd mercutio: oh? up_the_irons: i think i'm gonna load this up on one of my microcloud blades Webhostbudd_: give it a shot mercutio: smartos is doiugn things new and fancy
like they want you to network boot up_the_irons: even more reason to finish that box and haul it to the d/c Webhostbudd_: mercutio: which is great mercutio: and have all the storage on a machine running under zfs Webhostbudd_: imagine how easy it is to deploy nodes mercutio: which isn't a new concept Webhostbudd_: nope
don't the big guys do it? mercutio: people have been doing such with esxi etc for a while Webhostbudd_: vmware
yea mercutio: but like they're working towards "well thought out" ways of doing things Webhostbudd_: which i like mercutio: but also allowing a steeper learning curve
and doing things diffrently
like they're not trying to just support a whole lot of legacy ways of thinking Webhostbudd_: nope mercutio: they're trying to streamline things etc Webhostbudd_: i really appreciate the effort for a solid, opensource hypervisor
linux really doesn't fit the bill
well, the distros of today don't
illumos is a very solid kernel though
hardware support is probably shaky though
outside of your common server hardware mercutio: it's better than esxi for hardware support up_the_irons: i would only want to run common server hardware ;) Webhostbudd_: which is outstandingly sad up_the_irons: the uncommon stuff is expensive... mercutio: up_the_irons: one of the older hp's was broken a while back Webhostbudd_: hp's make me cringe mercutio: cpqarwy o rwhatever scsi controllers
cpqary3 up_the_irons: if it'll run on an Ivy Bridge 3.4GHz + plain ol' SATA + Intel GB, then that's all i need :) mercutio: http://www.greenviolet.net/articles/2011/09/21/openindiana-151a-and-hp-smart-array-controllers.gv Webhostbudd_: anyone here in the market for a sas expander? mercutio: up_the_irons: yeh it'll do that up_the_irons: mercutio: yeah i didn't figure that basic of hardware would be a challenge mercutio: up_the_irons: you need nehalem or whatever or newer cpus for kvm under illumos up_the_irons: mercutio: well sure, u need that for kvm in general Webhostbudd_: yes mercutio: well the hp thing would effect quite a lot of older hardware
up_the_irons: nah Webhostbudd_: ehhhh
nah up_the_irons: wut mercutio: up_the_irons: e5430 won't run kvm on illumos Webhostbudd_: i run kvm on adm phenoms up_the_irons: mercutio: oh? why? mercutio: because they're relying on some newer intel feature up_the_irons: ah Webhostbudd_: yes mercutio: uhh it hikn it's EPT up_the_irons: interesting mercutio: i can't remember what it was Webhostbudd_: that sounds correct -: up_the_irons googles Webhostbudd_: they don't support amd at all do they? up_the_irons: oh, extended page tables? mercutio: web: that was the case last i knew up_the_irons: no amd support?! Webhostbudd_: nope mercutio: up_the_irons: that's what it stands for but i don't know if that' the case
or if it was anothe rfeature up_the_irons: k mercutio: up_the_irons: someone may fix the amd stuff sometime
but smartos developers focused on intel up_the_irons: i'm loving my AMD 6212 system... Webhostbudd_: Currently KVM on OpenIndiana only supports recent Intel VT chipsets. up_the_irons: Webhostbudd_: ah gotcha Webhostbudd_: bulldozer opterons huh? up_the_irons: yup Webhostbudd_: do you get 16+ cores on those? up_the_irons: the 6212's have 8 Webhostbudd_: oh up_the_irons: but two sockets on my board, so 16 altogether mercutio: Finally, there is no support currently for AMD SVM. This is not a value judgement of AMD's technology, but rather a reflection of limited engineering and testing resources. (In the spirit of full disclosure, it should be said that the sponsor of illumos-kvm, Joyent, is an Intel-funded company -- but the lack of AMD support reflects only engineering prioritization and lack of testing infrastructure; AMD SVM support would be most welcome should someone in the Webhostbudd_: it sucks that those 8 cores get beaten so badly by 4 intel cores up_the_irons: Webhostbudd_: my cpus: http://i.imgur.com/Ed6nw.png
Webhostbudd_: in benchmarks or real world operation? Webhostbudd_: actually more in real world up_the_irons: cuz i'm finding them to be quite awesome Webhostbudd_: but a lot of server stuff is multithreaded and integer intensive so bulldozer should be a good fit up_the_irons: yeah mercutio: you've also got context switches to worry about
and having more cores can cut down in that when virtualising up_the_irons: exactly
context switching is super high in virt mercutio: i've never seen any good benchmarks for high load virtualisation
of multiple diff applications Webhostbudd_: no, and that is a problem up_the_irons: i find that benchmarks are like job interview questions; they look good on paper but do not reflect how the thing will behave in real life situations Webhostbudd_: exactly mercutio: up_the_irons: well it depends if you find the right benchmarks.. Webhostbudd_: you can also find the right interview questions mercutio: like you cna easily find game benchmarks to know if a video card will be a good fit for you up_the_irons: mercutio: maybe i'm looking at the wrong stuff then mercutio: but finding cpu benchmarks for virtualisation ... yeh.. Webhostbudd_: mercutio: that's pretty hard to say though up_the_irons: mercutio: yeah, games are specialized, cpus are generalized mercutio: mm Webhostbudd_: that's actually the problem with game benchmarks though mercutio: i'd look at kernel compile time benchmarks probably Webhostbudd_: unless you always play the same games... mercutio: but they're still not going to be accurate
webhost: you're comparing video cards to see what would be "good enoguh" for you :) up_the_irons: mercutio: kernel compile time might be a good overall heuristic, it uses disk + cpu Webhostbudd_: compile benchmarks have a lot of floating point though right? mercutio: webhost: nope up_the_irons: i would imagine
no? mercutio: why would compilers be floating point?
video encoding etc would be
but compiling is gnerally memory/integer heavy Webhostbudd_: see, i would assume that mercutio: if you add 2.5 + 3.5 on a compiler using float point numbers
the compiler only has to generate the assembler
that doesn't require it actually doing the maths. Webhostbudd_: oh im not talking about that up_the_irons: mercutio: perhaps at the optimize phase there is fp (but not lexer, parser, etc...)
*optimization Webhostbudd_: yes
this is what i always thought mercutio: well, if you multiply by 1.5
it probably does a shift
and adds it to the number
rather than actually doing it in floatin gpoint Webhostbudd_: im not talking about arthimetic in code at all mercutio: hmm
there's not that much optimisation work these days.
there's some, but it's not really as fancy as people often seem to believe. Webhostbudd_: obviously
fancy stuff breaks mercutio: heh Webhostbudd_: people like to believe a lot of cool stuff is at work mercutio: it's mostly stuff like cache line optimisation that is easier for compiler to do than by hand Webhostbudd_: mhmm mercutio: and reordering to prevent stalls. Webhostbudd_: yep mercutio: i'm still surprised by how big the linux kernel is Webhostbudd_: freaking massive mercutio: i'd like a kernel < 1 megabyte compressed really
i have one with lots of stuff disabled and it's still 3.4mb compressed
back with linux 2.1 it wasl ike 500 or 600k compressed Webhostbudd_: lol freebsd kernel so small mercutio: how big is freebsd kernel? Webhostbudd_: that's not real
bullshit
630k
really mercutio: openbsd is 5.1 mb
did freebsd go to modules? Webhostbudd_: i do have a lot of stuff disabled in my kernel
freebsd does have modules
but i also removed a ton of built in device support mercutio: openbsd isn't compressed
nor modular Webhostbudd_: well
my freebsd kernel doesn't even have ufs support mercutio: oh Webhostbudd_: i mean, i compiled everything out mercutio: it can't be zfs with that size
what fs are you using? Webhostbudd_: zfs doesn't build directly into the kernel
and zfs is only 120k mercutio: oh Webhostbudd_: it's not huge mercutio: that's good up_the_irons: nice Webhostbudd_: sec mercutio: i've got this thing against bloat :) Webhostbudd_: alright, i was massively mistaken
=p
2mb for zfs.ko
lol mercutio: heh
i thought zfs was big :)
it'll have support code etc Webhostbudd_: well, i knew it was big, i thought most people just overestimate mercutio: heh
zfs is a bit of a memory hog too
it's nice ram got cheaper :) Webhostbudd_: oh god yea up_the_irons: zfs is quite a memory hog Webhostbudd_: zfs is fast with lots of ram though
blazing
i love that arc cache mercutio: weird this page says kvm just needs vmx Webhostbudd_: which kvm? mercutio: illumos Webhostbudd_: linux kvm probably
oh
maybe they added support recently? mercutio: ARC Size: Current Size: 3598 MB (arcsize) Target Size (Adaptive): 3598 MB (c) Min Size (Hard Limit): 638 MB (zfs_arc_min) Max Size (Hard Limit): 5111 MB (zfs_arc_max) Webhostbudd_: and maybe that OI wiki is out of date jdoe: "memory hog" is another word for "aggressive caching" ;) mercutio: https://github.com/joyent/illumos-kvm/ ***: fink has quit IRC (Quit: fink) mercutio: it goes real slow with 2gb of ram Webhostbudd_: so wait.... that thing works on amd!
oh mercutio: what? Webhostbudd_: derp mercutio: svm is amd Webhostbudd_: yea mercutio: that page makes installing kvm really hard
it's actually simple on openindiana Webhostbudd_: just use smartos =p mercutio: EPT is supported on E5430
maybe it does work on earlier cpus? up_the_irons: interesting mercutio: what's this omnios Webhostbudd_: people love illumos
jeese up_the_irons: anyone know of a good iscsi target for linux? i see iscsitarget and open-iscsi (and i actually don't know of open-iscsi can be a target...) Webhostbudd_: no idea
god, i can't believe illumos is built with gcc 3.4.3
that just feels super old now