| |
| |
| |
|
Page: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Comments:
<0> ?
<1> arpf: IP6 is supported at the kernel level, and in standard routing tools. Distribution should be relevant
<2> DaveHowe: I prefer the way that raid handles redundancy.
<3> fbset: You might get away with it
<1> s/relevant/irrelevant/
<3> fbset: If you have no choice, I'd say it might be worth a try, and since you have backups anyway
<2> I run lvm so I can resize my partitions on the fly, and backup my database servers (nearly) on the fly.
<4> i think so atompngn
<3> BTW, SWAP on software RAID5 isn't the best idea, performance wise
<5> arpf: in any case, it doesn't matter. any sane setup would support both ipv6 and ipv4 until there were few enough v4 users to worry about... and good luck with that, I don't expect to live long enough to see it :)
<0> PolarWolf, all i need to do is ... format /dev/hda2 as SWAP ? then remove hdb2 from md1
<0> then... add hda2 to md1 ...
<2> yeah, swap on raid itself is stupid.
<2> if you have multiple swap partitions on multiple spindles, all with equal priority, the kernel will automatically distribute the load anyhow.
<1> arpf: I'm saying that distribution is irrelevant, because they are all equally capable of doing IPv6
<6> DaveHowe: so youc an mirror on lvm huh? well that's interesting
<5> libolt: sure
<5> libolt: and EVMS makes it look sick
<2> libolt: yeah, but you can't say do striping + parity
<4> DaveHowe i'm copirating with siemens and juniper know and i think guys there are motivated for the project DaveHowe
<3> fbset: I can't really tell how things are set up at your end (sleepy)
<5> Strider: this is of course true - but he is doing mirroring
<6> DaveHowe: well maybe I'll look into it sometime, my current setup works fine
<6> Strider: ah like say raid 1 + 0?
<2> libolt: no, like RAID 5.
<0> ok PolarWolf thanks alot for your help *nite*
<3> fbset: But I dunno man
<2> 1250284032 blocks super 1.0 level 5, 128k chunk, algorithm 2 [5/5] [UUUUU]
<6> ah right, raid 10 is mirroring and striping, what I run on this desktop
<5> it does 1+0 fine, but not 5 - because it can't use one redundancy segment to cover 2 (or more) data segments
<3> fbset: You have a software raid5 underneath a partition table which includes SWAP which have filesystems on them
<0> PolarWolf, well disscussing an issue always helps
<2> yeah, swap on top of raid 5 is stupid.
<3> fbset: The annoying part about raid5 is that you really have no idea where your data is. It's spread out over the disks
<2> my disk set has 6 spindles.
<2> PolarWolf: right, but in general you don't really care where it's physically located anyhow.
<6> I never use swap on top of my software raid setups, I just leave a little extra space on each disk to dedicate to swap
<2> as long as it's fast and safe. :)
<3> fbset: So if you just axe SWAP, you might actually be touching the end of your volume at all
<3> err, not touching, too
<0> PolarWolf, ohh then i might wornder why the raid go broken in the first place... it was made like this before aswell
<0> got*
<6> hrm I guess I'll have to eventually throw together a system to play around with LVM and EVMS in more detail
<2> fbset: are you running lvm on top of RAID?
<3> Strider: correct
<2> ok, then it's no big deal.
<5> Strider: hmm. I have a funny feeling higher levels of raid are supported in lvm2 and evms. hold on...
<2> just stop using the swap partition, use lvmtools to remove the partition, grow the other partition, and expand the filesystem.
<2> boom, raid gone.
<3> Strider: No, I mean correct on the not caring about where my data is :)
<2> and you don't need to down the system to pull it off either.
<2> ahhh
<3> By the looks of it, this dude doesn't use LVM
<0> PolarWolf, it's raid 1
<0> their is no parity here
<0> only mirroring
<3> fbset: Okay, then you might get lucky
<4> DaveHowe i have a friend who worked on a same project (thesis) and manage to find a solution for the encapsulation of ipv6 over atm and i think that juniper routers are more and more supporting ipv6 so what's the problem?
<2> arpf: what's hard about stuffing ipv6 over atm?
<3> fbset: Just reroute the swap to somewhere else (swapoff/swapon)
<4> raelly?
<2> (never mind the fact that ATM in an IP only network is annoyingly lame)
<5> arpf: no problem - but until at least 50% of your expected user base supports ipv6, you would be insane to discard v4 from a real world server
<7> my understanding of the IPv6 problem is the cloud of M$ devices at the edge of the network that can't/won't speak it
<3> fbset: Then axe the partition containing the swap, and make a new one which is smaller...though I really wouldn't swap on raid volumes
<3> IceKarma: My understanding of the ipv6 problem is that noone offers it, really
<5> arpf - and ATM sets up virtual circuits. you can run anything you like over them - from SLIP upwards
<3> IceKarma: chicken and egg problem, I say
<7> DaveHowe, eh, for a while I was running most of my services IPv6-only, IPv4 clients could still talk to them fine
<2> DaveHowe: the problem lies in actually making 'em many-point ot many point
<3> As long as I can't go to my regular household broadband ISP to request ipv6 connectivity, ipv6 is a dead duck
<2> <-- spent 2 years working in the ATM industry.
<5> IceKarma: only via an ipv4->v6 gateway. v4 only clients can't locate or address v6 hosts
<7> DaveHowe, that being said, the -box- was IPv4-connected
<2> ipv4 will never happen until it's an economic nescessity to do so.
<2> right now, it's not.
<2> it's just for geeks like us to play with.
<3> Anyway, I'm really off to bed now
<5> ok, EVMS supports RAID-5 regions. not sure if it supports higher redundancies yet
<4> ipv4 is dead manifucters have no choice guys
<4> even with nat
<3> ipv4 address depletion is a myth
<2> arpf: it's not dead yet, and I wager it won't be fore at least another 5 to 10 years, minimum.
<0> PolarWolf, then axe the partition containing the swap ? how, with ?
<2> which is well beyond the planning horizon for most kit.
<5> arpf: oddly enough, I heard that same bleat about a decade ago, which was when I started playing with ipv6
<3> fbset: I'm not here anymore :)
<0> :)
<0> ok
<2> fbset: once you use swapoff to unmount the swap, you can do whatever you like with that disk space.
<7> my prediction is that we will be completely out of IPv4 address space and really feeling the pain on that before any major switching happens
<3> Just re***ign a couple of A cl***es and start utilizing the holes in the current ***ignment
<2> fbset: but adding it to your existing drive is iffy.
<3> There, problem fixed
<4> me too IceKarma
<2> IceKarma: and given current ***ignment patterns, I don't see that hapening for a while except in asia.
<4> so what's the solution DaveHowe?
<2> I predict asia iwll go ipv6 long before the west does.
<5> if that - we may find that ipv6 becomes more comon at the "consumer nodes" but that servers stay solidly rooted in v4-space
<2> DaveHowe: I can also see ipv6 becoming the next big thing for m***ive server farms like google and what not
<8> DaveHowe: highly likely as more and more consumer crap becomes IP aware.
<4> don't forget UMTS guys!!
<2> the problem is that most ISPs will try to ***ign people a single ipv6 address rather than a /64 or /72
<2> and destroy network portability
<5> wibble - EVMS supports raid-6
<5> glods, but that must EAT cpu time
<2> DaveHowe: heh.
<2> IIRC, RAID-6 uses basically FEC no?
<4> and this problem is clear if we use 3rd generation networks all over the world!
<5> never ever seen that done outside of hardware
<5> Strider: RS encoding, same as PARsets
<2> ahh, Reed Solomon
<7> Strider, I think that IPv6 will place even more of a memory demand on routers than a full IPv4 routing table does today
<7> Strider, whether it's carrying around a bunch of /64s or /72s or /128s
<2> IceKarma: there's a reason why it's impossible to get a newly ***igned portable subnet smaller than /19 now.
<7> Strider, that's funny, the limit was /19 the last time I got a portable subnet, ten years ago
<5> yeah. BGP ****s, and a full BGP table is stupidly huge
<2> I wish I had been smart enough to get a portable subnet.. *sigh*
<2> not that my ISP would allow me to advertise it.
<7> Strider, this was for work, at the time
<2> IceKarma: yeah
<7> Strider, and we had to agree to give back our nasty twisty mess of cl*** Cs
<2> DaveHowe: it will get only bigger.
<4> but IceKarma processors are becoming more and more performant and meories more and more huge
<2> arpf: yes, but it's not scaling up exponentially
<7> arpf, that isn't supposed to be a challenge, though ("let's see how much RAM and CPU I can chew up!")
<7> like credit card limits... it's an upper limit, not a challenge ;)
<5> arpf: yeah, but BGP doesn't scale; now that we have lost the EBONE BGP paths have gotten noticably longer in europe too :(
<2> nothing scales well enough for the core routing.
<4> i think that guys at juniper didn't invest million of dollar developping router ready for ipv6 to play with??!
<2> efficient routing is a fundamentally tough problem..
<2> arpf: beats me, when I worked for Nortel we were working on a terabit router that never saw the light of day.
<4> DaveHowe there's no problem with BGP you can see the new rfc for ipv6
<5> arpf: cisco ipv6 support ****s. 95% of the world runs on ciscos. your point is?
<2> DaveHowe: thank god most of the core network doesn't run on cisco though
<4> :)
<2> your choices are basically ISIS or BGP
<5> Strider: doesn't run on juniper either though. still mostly foundry kit, isn't it?
<2> DaveHowe: not sure, honestly.
<5> Strider: EBONE was mostly foundry, before it went dark
<5> Strider: don't know about the american core(s)
<5> haven't been subbed to nanog in ages now
<5> arpf: there's a new one? haven't seen any this century. hold on while I google
<4> but DaveHowe we can use ipv6 into and AS (atonemous system) and tunneling ipv6 into ipv4 to connect 2 AS and the problem of BGP is avoided?
<5> arpf: you mean push AS routing into the conventional routing table in the hope that can cope? you *do* know why the AS system is there, yes?
<4> yes
<5> then you must know that trying to do a pure routing table mapped onto even ipv4 address space would be insane
<5> it has to be abstracted at the AS layer to make it work
<2> remember about 5 years ago when an ISP accidentally polluted the core routing tables, making their T1 look like the best link on the net?
Return to
#linux
or
Go to some related
logs:
touche
#c++
#php
#javascript
Cryptographicexception Key not valid for use in specified state
Lafejn
gnomebaker plugin handle is not installed
#teens
#java
rdragon lebanon
|
|