mercutio: 7.1 gb in the last month
***: mjp has quit IRC (Read error: Connection reset by peer)
mjp has joined #arpnetworks
Webhostbudd has quit IRC (Quit: Leaving)
up_the_irons: beta VM guys should re-test disk I/O. I've disabled my munin graphing temporarily (turns out it was munin going crazy, not cacti)
munin will wind down in about 5 minutes
and so now there shouldn't be big load spikes
-: up_the_irons leaves the data center
***: Ehtyar has joined #arpnetworks
up_the_irons: w00t, sponsored reddit link: http://www.reddit.com/comments/11axe2/arp_networks_offers_freebsd_vps_services_with/
i can haz upvotes?:)
toddf: are you compiling something or doing something heavy on your beta VM? (not a problem, just wondering cuz i saw it taking a bit of cpu)
***: jnq has quit IRC (Quit: WeeChat 0.3.0)
up_the_irons: anyone know if a munin master at 2.x can be used with munin nodes that are still just 1.x ?
i want to move to 2.x for performance on the master, but i'd hate to have to also update all my nodes
josephb: up_the_irons: I've always had master and nodes on same train. It should work though.
up_the_irons: josephb: roger
josephb: up_the_irons: tun/tap support, anything special needed?
up_the_irons: josephb: nope
josephb: pretend it is a bare metal machine that's all yours
josephb: ta, just trying to get ipip or gre tunnel going back to home so my vps can monitor internal things
up_the_irons: josephb: shouldn't be an issue. i see gre traffic here and there, so i know other people are doing it too ;)
unfortunately, i must dual boot into Windows now (radio programming)...
-: up_the_irons walks over to the dark side
mikeputnam: upvoted up_the_irons
***: LEMONed has joined #arpnetworks
LEMONed has quit IRC (Quit: !stfu)
toddf: up_the_irons: abusing it to test opensmtpd .. I think its wedged .. resetting
I started a test on friday, I guess it kinda went bezerk.
UVM: pid 30715 (procmail), uid 547 killed: out of swap
qbit: doh
toddf: yeah, until 'recent' commits this weekend, if you did 'echo "|sleep 60" > .forward' and delivered a bunch of tiny mails you'd end up with an out of fd exhaustion scenario where it falls flat on its face ...
local deliveries happened 'quickly' in any prior tests, but since I'm delivering mail via procmail and piping it through a dspam filter .. it takes a few seconds and the more that pile up, the slower it goes, circle of death ensues, etc ;-)
now there is a std default MDA limit per user and global for the system, to be tunable after the upcoming release of opensmtpd
qbit: wew
i am still on a mission to find a usable android emial client that isn't gmail
stupid conversationviewbeingawesome
toddf: qbit: 'usable' is in the eye of the beholder. ever consider writing one if you're searching in vain?
qbit: toddf: no doubt.. but building a java app of that scale is out side of my knowledge-time threshold.
even hacking on existing open source mail clients :P
***: dj_goku has joined #arpnetworks
dj_goku has quit IRC (Changing host)
dj_goku has joined #arpnetworks
plett: toddf: There is a reason why the traditional MTA design was to queue emails at all stages and have a queue runner to process them later
toddf: plett: are you suggesting smtpd does not do it that way?
plett: it does, infact, queue all mails and then process it later, and there is a mda (mail delivery agent) process that you might consider a queue runner to process them later. it simply did a 'forkmda()' call without bounds. that is now fixed ;-)
plett: toddf: Ahh. I didn't know opensmtpd did work that way. I read your initial comment to be that it spawned deliver processes for each email without considering how many other deliveries were already happening
toddf: apparently I was the first one to test with a procmailrc that took more than <1s to deliver ;-)
milki: ol
l
***: jbum has quit IRC (Quit: jbum)
jbum has joined #arpnetworks
RandalSchwartz has joined #arpnetworks
-: RandalSchwartz waves for no apparent reason
toddf waves back
***: warex has joined #arpnetworks
warex: Is there a link on how to configure the parameters of the network device? (I'm manually installing OpenBSD)
toddf: man hostname.if(5) and if you don't have an openbsd host handy see http://www.openbsd.org/cgi-bin/man.cgi?query=hostname.if
RandalSchwartz: *everyone* should have an openbsd host handy. :)
twobithacker: bah, just use FreeBSD and get the best bits of OpenBSD :)
RandalSchwartz: there's also the http://www.freebsd.org/cgi/man.cgi
which has *all* the manpages
nice resource
warex: I'm sorry, I mean the parameters from arpnetworks to configure my network device (ipv6address, ipv4address and else)
brycec: warex: all of that's in your portal
Notably, under "IP"
warex: oh... sorry! thanks!
***: chmod2 has joined #arpnetworks
chmod has quit IRC (Ping timeout: 246 seconds)
HighJinx has quit IRC (Quit: Computer has gone to sleep.)
chmod2 has quit IRC (Read error: Connection reset by peer)
chmod has joined #arpnetworks
Webhostbudd has joined #arpnetworks
_mnathani_ has joined #arpnetworks
HighJinx has joined #arpnetworks
warex has quit IRC (Ping timeout: 245 seconds)
dj_goku has quit IRC (Quit: leaving)
warex has joined #arpnetworks
warex: is it needed to disable mpbios so OpenBSD can start its em0 networkservice? I'm trying a fresh install but em0 sends a watchdog watchdog timedout and freezes just when trying to start the network services
toddf: yes
***: HighJinx has quit IRC ()
HighJinx has joined #arpnetworks
Ehtyar has quit IRC (Quit: Never look down on someone unless you're helping them up.)
up_the_irons: mikeputnam: tnx!
toddf: ah ok
qbit: what about k-9 mail?
that's what i use
mike-burns: I also use K-9. It's certainly no mutt, but it's the best I've found.
qbit: up_the_irons: i wish i could set colors for specific headers
***: mjp has quit IRC (Remote host closed the connection)
qbit: and thread things by converstaion
CaZe: qbit: You mean the three long bolts?
qbit: conversation too
CaZe: ya
CaZe: I think I tighten them quite a bit tighter than he recommends.
I never did describe the actual alignment process, did I?
qbit: nope :D
CaZe: Well, like I started saying, you have to first think about why being out of alignment causes the axle to bob up and down.
Do you know why that happens?
qbit: lol, because it's not aligned. :P - same concept as a bent axle
CaZe: No, the axle is straight.
At least, it should be.
If it isn't, then you need a new axle.
qbit: i just mean in concept - not sure how else to explain it other than that
CaZe: And a bent axle doesn't bob up and down parallel to its shaft anyway.
qbit: oh
then i don't know :D
CaZe: Well, as Doug describes, the inner burr isn't perfectly round.
It's slightly oval.
Do you have your Pharos in front of you right now?
qbit: yep
***: mjp has joined #arpnetworks
qbit: and if i spin the axle when it's upside-down the axle bobs
CaZe: Well that wouldn't happen if it were aligned, because even if the inner burr is oval, if it were perfectly centered, then the two "far" points of the oval would be scraping the outer burr, counterbalancing one another.
qbit: so it bobs because of the oval-ness of the inner burr?
CaZe: It bobs because it's out of alignment.
If it's out of alignment, then that means the inner burr isn't centered.
qbit: so the burrs are hitting when it's bobbing
CaZe: The burrs are always hitting when you have it upside down.
The burrs are the only thing that keeps the axles from sliding out of the apparatus and falling on the floor.
s/axles/axle/
qbit: makes sense
CaZe: Having it upside down puts the burrs at "zero", as Doug calls it.
qbit: k
so you loosen it and spin the axle to get minimal bob
CaZe: So when the inner burr is off center, one of the "far" points of the oval is scraping the outer burr, without being balanced by the other "far" point of he oval.
Keep in mind that the outer burr doesn't have straight walls.
It's a cone.
qbit: right
***: chmod has quit IRC (Ping timeout: 246 seconds)
CaZe: So as the far point of the inner burr comes into contact with the outer burr, it wants to try and push into it.
But since it's a cone, to the inner burr, it looks like a ramp.
So the inner burr basically creeps up the "ramp", bringing the entire shaft assembly with it.
qbit: so shift the outer burr to minimize "ramp"
CaZe: It's basically the way that the far point of the inner burr pushes through the wall of the outer burr.
It seeks the path of least resistence, and that means creeping up along hte wall of the cone, where the diameter of ther outer burr is larger.
Well the ramp is always there, you just want the far ends of the oval to be balanced against one another.
mercutio: up_the_irons: you around?
CaZe: qbit: It might be easier to visualize if you take out the burrs and hold them relative to one another in space.
qbit: think i have a pretty good visualization
***: chmod has joined #arpnetworks
Ehtyar has joined #arpnetworks
CaZe: Actually, let me revise that.
I just found my notes.
I was describing to you an earlier version of my theory.
There doesn't have to be any balancing.
In fact, the inner burr doesn't have to be an oval.
It could be (and probably is) an ellipse.
Where one of the far points never touches the outer burr.
As far as our model is concerned, the inner burr can be modeled as just a single stick protruding perpendicularly from the shaft.
If you imagine the stick to be a piece of chalk, then it draws its path on the surface of the outer burr.
qbit: k
CaZe: When it's perfectly centered, the path is a level circle.
qbit: right
CaZe: When it's off center, the path it traces is an ellipse.
It climbs higher along the cone, then drops down to a lower point.
http://csep10.phys.utk.edu/astr161/lect/history/conic.gif
Compare circle and ellipse in that diagram.
The cone is the outer burr, of course.
qbit: aye
CaZe: The motion of climbing higher on the wall of the cone and then dropping down again is what causes the bobbing motion of the axle.
qbit: yep
CaZe: When it's centered, the path is a level circle that has a constant height along the cone, so there is no bobbing.
qbit: yerp
CaZe: If you do as Doug says, and leave the three long bolts loose, and just rotate the handle, then it self-aligns.
qbit: that's how i did it last time
tightened ~ to his spec..
but it's wobbled loose
CaZe: Since the outer burr is free to move around, the inner burr just sort of pushes the outer burr into place, since that's easier than climbing the wall and moving the entire shaft up along with it.
Well you probably want to tighten it more.
But the problem when you do that is that overtightening actually puts it out of alignment.
And I think that's why he tells people not to overtighten.
Now, we have to investigate *why* overtighening puts it out of alignment.
qbit: because the middle plate moves the outer burr
* moves it's angle
CaZe: Yeah, sort of.
The appartatus isn't a perfectly rigid structure.
As you tighten, things bend slightly.
Enough to put it out of alignment.
***: chmod has quit IRC (Ping timeout: 246 seconds)
qbit: k
CaZe: But if you only tighten one bolt at a time, and check the alignment before you do anything else, then you *know* for sure how it was distorted.
And fortunately, he only used three bolts, which makes this process easier.
If he had use four, then it would not only be a longer process, but more complicated.
So when you tighten a bolt, it pushes the steel plates closer to one another on that side.
It has the effect of tilting that side of the outer burr towards the handle.
qbit: which may or may not be a good thing.. depending on the ellipsieness
CaZe: Well, as far as alignment goes, what effect does tilting the burr have? And how could you correct it?
***: chmod has joined #arpnetworks
qbit: seems if you loosen the bolt by the "ramp" and tighten the others .. it should reduce the ramping
CaZe: Yes, but then you have one bolt loose.
qbit: right - but i don't mean all the way
CaZe: And how do you know that tightening the other two bolts won't put it out of alignment in another direction?
qbit: by respinning
CaZe: But that won't work once it's torqued down.
That's the whole point of overtightening - so that *won't* happen.
Otherwise you'll be realigning every week.
Tightening a bolt has the effect of lifting the side of the outer burr that the bolt is on upwards, towards the handle.
As far as the "ramp" goes, it's the same as moving the outer burr closer to the inner burr.
Since the ramp is being lifted, the "stick" representing the inner burr is now touching a lower part of the cone.
Actually hold on, let me visualize this to make sure I have the direction right.
Maybe I should draw a picture.
***: chmod has quit IRC (Ping timeout: 246 seconds)
qbit: heh
right on
***: chmod has joined #arpnetworks
CaZe: Hmm, I have to go make dinner.
Anyway, just tap the burr on the opposite side of where you tightened the bolt.
Since tightening that bolt reduced the clearance, shifting the outer burr in the direction of that particular bolt will increase the clearance.
And just do that for each of the bolts. (You'll have to tap harder as you progress through the bolts.)
It's possible to tap it too much, and it's tricky to tell when you've done so.
But since the configuration of the three bolts is a triangle, pushing the outer burr too far when you're tighening one of the bolts is the same as putting it out of alignment with respective to the two other bolts.
And you can correct for that when you visit the two other bolts.
qbit: ^
qbit: rad
thanks :D
have a good dinner!
Webhostbudd: up_the_irons: you there?
up_the_irons: Webhostbudd: yes, but busy
Webhostbudd: oh
alright
up_the_irons: quick questions, no problem; need me to do something, gotta wait :)
Webhostbudd: I'll just submit a ticket then
up_the_irons: roger
Channel poll: raw backup space (rsync) to be available for $0.20 per GB, good price? A bit cheaper than Linode and 30% cheaper than RootBSD
Webhostbudd: hmmm
that sounds pretty good
$0.20 per GB per month?
up_the_irons: yeah
Webhostbudd: yea
i like that
up_the_irons: so if you had a small plan ($10) with 5GB disk, you'd pay only $1 more for enough space to fully rsync all that
Webhostbudd: but im guessing you can rsync just individual files too
no one is going to sync their entire fs....
at least, i would hope not
up_the_irons: yes, that's the idea, you run rsync within your vps
i think i'd build that backup host with FreeBSD + ZFS mirror. Initially with just a 2 disk mirror to get it off the ground. I already have a blade ready for it..
chroot home dirs, like ryk suggested
Webhostbudd: yea
definitely
ryk: up_the_irons: you have a blade server?
thought you use supermicro twins
or super twins or whatever they're called now
up_the_irons: ryk: i do have a blade server, not the twins
food time
-: up_the_irons grabs grub
Webhostbudd: lol
CaZe: up_the_irons: I'd be up for it.
qbit: hot
mercutio: up_the_irons: raidz probably makes more sense?
up_the_irons: i wonder if it's possible to do some kind of lvm snapshot copy type backup...
which you could "recover" from if you wanted to
the problem with mirror is that if you want to grow it you'd have to add two more disks, or add a raidz on .. as zfs doesn't allow changing from mirror to raidz
ryk: raidz would not be a good choice for this, imo
1) it's backup
2) raidz is not easily expandable
now, you can keep adding raidz devices to your zpool, but you have to add an entire raidset every time you want to expand it.
mercutio: mirroed you have to add a pair each time
raidz you have to add 3 disks every time
err 3 or more
ryk: and of course you're going to want to do more, because eww 3-disk raidz?
so that gets expensive.
mercutio: what's wrong with 3 disk raidz?
***: jbum has quit IRC (Quit: jbum)
ryk: what's wrong with a mirror vdev?
mercutio: 3 disk raidz gives you twice as much storage space as mirror
it also gives you faster write speeds
ryk: but if you know up_the_irons , you know he's a raid 1- guy
*10
mercutio: it's a bit different for backup space to virtual machine storage
ryk: there's no requirement for speed, here. i'ts backup space.
mercutio: well yaeh, most of the time there isn't
but if lots of people cron at the same time their backup..
and someone tries do a read at the same time
or noties their backup taking ages
raidz gives higher write bandwidth
ryk: i mean, you're right -- if it had to be parity raid, i would choose raidz over raid5
***: HighJinx has quit IRC (Quit: Computer has gone to sleep.)
mercutio: but yeah, even for normal performance generally speaking, i'd say that 4 raidz3 set's stripped would be competitive with raid10 with 12 disks
ryk: raidz3? or 3-disk raidz?
mercutio: although when getting to 12 disks the chances of dual disk failure increases
ryk: i thought raidz3 was super-redundant
mercutio: 3 disk raidz
sorry, confusing terms a little :)
i don't think there's a raidz3
there's raidz2
ryk: there is, now, it's new-ish
mercutio: generally speakign above that i think people say you should have more than one data set or more than one raid volume
err more than one pool
i'm confusing terms again
anyone the rational was if you have 3 2tb hard-disks with raidz you get 4 tb of disk space, if you have 4 2tb hard-disks with 2 pairs of mirrored drives you get 4tb of disk space
with 6 disks.. raid-z gives 8 tb .. raid mirrored gives 6 tb
i suppose it's not a big diff
it'd be better if you could grow storage nicely
like just chuck another 2tb in and rebalance
ryk: so, having never used raidz, i haven't measured write speeds. you're saying write is faster?
because that's contrary to conventional wisdom i would associate with parity raid
read is usually faster, write is slower due to the calculations
mercutio: calculations are minimal
bsaically it can write half the data to one disk half to the other
as well as the parity
so if you write 100 gig of data
you write 50 gig to 3 disks
whereas with mirrored raid you write 100 gig to 2 disks
an thus you write 150 gig versus 200 gig
milki: and read?
mercutio: read performance is similar for sequential
the problem with raidz is for lots of small reads
you reduce your seek performance, because say you have 128k stripe size per disk
then you'd have to read 128k off each disk
for even a 4k read
limiting seeks to that of a single disk
with backups though they're by far sequential
so it can just read huge chunks off each hard-disk
or write big chunks to each hard-disk
but y eah when people say about the overheads of raid parity calculations it's way less significant than aes (https, ssh, etc)
sure it may have been a significant overhead in 486 days
or more so in sparc days
err early sparc days
but it's not a significant concern.
aes is more likely to be a concern
if lots of people scp/ssh at once
but even then if you have multiple cores without hardware acceleration it should be fine anyway
also another thing to keep in mind with raid10 versus basically raid 50 is that most work loads cache the majority of their reads
but you can't really cache writes, you can buffer them...
but sooner or later you have to write them to the disk
whereas reads just can from memory a lot of the time
so for things like hosting web sites, it's mostly a "write load" not a "read load"
because of appending the log file with updated accesses etc
although it's also mostly small writes
but if you can't keep your most visited documents in memory, and are frequently going to disk, then either you're starved for memory, or you have a huge dataset
and if you have a huge dataset and it's text files, it may make more sense to compress your memory cache
apache traffic server is an example of a proxy server that can compress it's memory cache
i don't know why it's not used more often, it's faster than varnish or squid or ngnix
ngnix
***: warex has quit IRC (Ping timeout: 245 seconds)
ryk: apache traffic server?
-: ryk googles
mercutio: it was a yahoo project that went open source and was made part of apache's greater uhh
whatever it's called
it's multithreaded, high performance, under active development
CaZe: up_the_irons: Another thing that might be nice is a domain for customers.
You know, for those of us who don't care to get our own.
Those free dns services are a little annoying because you have to update the ip evey once in awhile, even with a static ip, or else they drop you for inactivity.
qbit: I'm investigating another possible reason for why tightening the bolts puts the burr out of alignment.
***: HighJinx has joined #arpnetworks
CaZe: qbit: It has to do with the type of nuts that are on the end, and how they grab the bottom plate as they're tightened.
If the long bolt is not perfectly straight through each plate, tightening the nut will straighten it.
But if the teeth in the nut have grabbed into the plate, then the position of the plate will shift with the nut.
I'm going to see if adding a washer will change anything.
But it'll have to wait until tomorrow.
qbit: right on
seems that if you gighten from the top - and keep the bottom in one position it owuld hav ethe sameish effect
CaZe: Maybe, but it still wanders a bit.
btw, I got rid of those four smaller screws holding thd outer burr.
And the springs, too.
qbit: interesting
whY?
HighJinx: cpu hs mounts?
CaZe: Because I think they just get in the way if you're planning on overtightening.
They certainly don't add anything.
qbit: true
CaZe: HighJinx: Coffee grinder.
HighJinx: ahh
CaZe: If you're going to tighten to the point that the springs bottom out, then there's no reason to have them.
And the only reason the four small bolts are there is to keep the spring in place.
Webhostbudd: has anyone had success with virtio on kvm?
errr
freebsd virtio modules on kvm
mnathani: I have been working on some web based DNS tools, if anyone would care to provide some constructive criticism: http://dns.winvive.com
staticsafe: mnathani: pretty array formatting :)
brycec: lol gotta love print_r ;)
mnathani: :-) staticsafe
staticsafe: I like it tbh, simple and clean
brycec: Odd... according to your site, mnathani, google.com isn't DNSSEC. That doesn't seem right
mercutio: webhost: i've had openbsd work with virtio on kvm
mnathani: I modeled some of the tools after http://www.dnsstuff.com/
mercutio: but openbsd takes code from netbsd not freebsd
brycec: mnathani: +1 for ipv6
mnathani: Thanks brycec
dont think google.com is dnssec yet
winvive.com is however
staticsafe: damn, really need to read up on dnssec
brycec: touche, google.com really isn't signed
You just figure... IT'S GOOGLE
-: brycec just found https://chrome.google.com/webstore/detail/dnssec-validator/hpmbmjbcmglolhjdcbicfdhmgmcoeknm too, for those who care
brycec: ah indeed you are mnathani http://dnssec-debugger.verisignlabs.com/winvive.com
-: brycec should get around to figuring out what's needed in signing his domains
mnathani: I have a bunch of tools like that over at my dnssec signed blog: http://mnathani.com/blog/dnssec-links/
*links to tools rather
brycec: What, using your own IP is too hard? Not knocking the idea really, just curious at the gap between "I can't be arsed to get a domain for $1.10/yr" and "I can't be arsed to type an IP address or edit my hosts file" [19:42:48] <CaZe> up_the_irons: Another thing that might be nice is a domain for customers.
up_the_irons: The rsync backup space is a neat idea. Critical question time - why would I use that space over Amazon's S3 RRS which is $.09/GB/mo? Besides saving on VPS b/w costs.
-: brycec already uses S3 RRS for backups now, every single server I admin, total about $16/mo
brycec: Aaaand before I'm over this dnssec kick - hey up_the_irons why isn't arpnetworks.com DNSSEC signed? :)
mnathani: here is another question, why are the arpnetworks customer resolvers not validating DNSSEC?
CaZe: Domain names are easier to remember than ip addresses.
jlgaddis: < CaZe> up_the_irons: Another thing that might be nice is a domain for customers. <-- you mean like the dns records for "your-name.cust.arpnetworks.com" that are created when you sign up?
CaZe: Ah, I didn't know about that.
But it'd be nice if we could change the name.
mike-burns: That's what name.com is for.
***: _mnathani_ has quit IRC (Ping timeout: 240 seconds)
mnathani has quit IRC (Ping timeout: 255 seconds)
_mnathani_ has joined #arpnetworks
mnathani has joined #arpnetworks