this is entire estimate is meaningless. v6 was designed to be rolled out incrementally. no one beleives that all the endpoints are going to be upgraded.
so if some of the major backbones start peering v6, thats a good definition for switching, but i seriously doubt its going to involve tens of billions of dollars.
the incremental cost of new larger customers being assigned v6 blocks instead of v4, and having to push it to the endpoints or put in nats? the dns servers (the only thing of any substance that was mentioned)...do the v4 roots cost billions of dollars a year?
what, are you kidding. thats exactly what one of those godless queers at the university would do to besmirch the name of our lord jesus christ.
why he deserves...hey, joey, get your baseball bat...dont ask any questions, just do it.
Re:That's Because "IT Doesn't Matter!" Anymore...
on
Security's Shaky State
·
· Score: 1
your statement concerning the symptom and the disease is key.
the problem isn't that we dont have enough people trying to clean up the mess. the problem is the mess. although key distribution is a difficult problem, the basic infrastructure needed to provide relatively secure distributed services has existed for almost 20 years. and its still not in common use. the idea that reading my mail can give a random person local administrative access on my workstation is obscene at best.
hiring people to try to make things sane given the state of the infrastructure is a losing battle...fingers in the dike. the real question is why as a community we dont put up with the incremental pain to put ourselves on firmer ground.
there is a reason why things tend to go as powers of two in computer architectures. think about it, it will probably come to you.
given that there is a general crossbar between the processors and the memory units in the amd processors, this probably doesn't need to be such a problem here.
but having a non power-of-two factor anywhere really screws alot of things up. even seen a 3GB ram? a 37 bit bus? a 27 element register file? (alright, forget this last one, x86 is a mess)
i agree with the other posters, there must be something better for these people to spend their time on.
you're absolutely right. by even admitting that there is a chain of relations between species (the fossil record), and then saying that its broken because it has holes, there is no way that person really holds the belief that all species were independently created.
wealth creation? i'm somewhat familiar with this capitalist fanboy term, but in what sense does emule/edonkey create wealth? it moves bits around. i'm sure its very nice.
some garbage collection environments let you associate a destructor with a given object or type of object. it still means that your connection will hang around until some point after there are no references, but at that point you can arrange to call close()
umm. it works the same as malloc. you get raw pages from the operating system, and you have to spend some of that space keeping track of whats free.
one (overly) simple approach is to keep a list of 'partially allocated' pages for each size of interest (2^n). in each page there is a freelist and free count. an allocation is thus two indirections in the common case. a completely empty page gets given back to the os, or put on a list of unassigned pages. multi-page objects are handled differently (always as a multiple of the page size)
there are of course many other approaches. the basic literature is very accessible if you're interested.
the solution to allocation inefficiencies due to fragmentation is well known. it doesn't require any substantial changes to the programming model. use a coalescing garbage collector. since you are moving objects it requires either use of mprotect() to trap stale references or lock the object against mutation during the copy, or the assumption that one can 'stop the world' while copying objects.
as usual the overhead of copying can be tuned by policy, and mitigated through the use of a standard generational implementation.
having just exposed a memory leak in ff 1.5 using svg w/ javascript, i really do wish they would have chosen to just gc the whole thing
from my limited understanding, the back end google clusters are designed to be fault-tolerant and somewhat redundant.
so i expect that the model is that components will keep dropping out until the unit is no longer sufficiently capable or cost effective, and then they would swap it out with a freshie.
heat is clearly an issue, which is one of the reasons why they needed industrial designers. the most effective solution would probably be to have smaller heat exchangers inside the box with internal air circulation, and require an external chilled water feed.
are you seriously trying to claim that cray occupies a position of marketing excellence?
from a business point of view its completely whacked. cray spends 4-5 years of time to build a machine, just to sell a very small few of them, throw almost all of the technology away and start over again.
and are you really trying to claim that any cray machine built in the last 10 years has particularily good mtti rates? the sv2 really was basically unsuable for the first two years after it was shipped. its still kind of a dubious proposition.
its not ironic at all. its a question of resources and volume. cray has a few very bright people (still, sort of). they are essentially a us government lab. they do a bad job, but its insane to think that 100 people can build and maintain several different supercomputer architectures.
a $300 opteron is almost always more effective than a $60000 X1 processor. they have alot of bright people too, and alot more of them.
the only reason that cray still exists is support for parallelism and the provision of high memory bandwidth systems. but even that niche is being eroded pretty severely. the xt3 communications chip runs at 3.5GB/s in each direction. it costs about $250 for cray to have each of them made. for the same $250 i can buy a mellanox nic that runs at half the speed
its no suprise that cray is using opterons. they actually got lucky by committing to amd early and having it turn out so well.
the real question is whether there is any more room for a cray at all. the commodity world moves so quickly. the xd machines (which they purchased) are really their best asset, but it still hard to justify that kind of margin for what is essentially a well constructed cluster.
i think you're a bit off the mark. normally as you say someone would get a cs degree, and do additional work in comp ling, then apply to grad school with an specific bias towards that kind of work.
however, if you really know thats what you want to do, there's alot to be gained by putting together an interdiscplenary degree based around your interests and getting it approved. it would mean you would be able to take more courses in linguistics and philosophy of language, or statistical methods, or logic, or whatever you think you need and apply them to your degree. it means as you suggest more work, because you probably have to develop a basic grasp of cs as well as all those other things. but if you're sufficiently motivated, why not? it may also help to get approval to take grad level courses in the areas you think are necessary, given the general paucity of undergrad level materials at that degree of specialization.
however, if it turns out that that isn't what you end up wanting to do forever, you have kind of painted yourself into a corner.
i also disagree about the job issue. if you're willing to accept a narrowed field, i can think of a few companies who would love to have comp. ling. people on staff. i've even seen job advertisements.
this person is setting a higher bar for themselves and committing to it.
you're right in some sense, the pressure underneath the plenum will force air through no matter what. there are however two problems. the first is that turbulence underneath the floor can turn the directed kinetic energy of the air into heat...this can be a real drag. in circumstances where you need to move alot of air, the channel may not even be sufficiently wide.
more importantly, the air ends up coming out where the resistance is less, leading to uneven distribution of air. if you're grossly overbudget and just relying on the ambient temperature of the machine room, this isn't a problem. but when you get close to the edge it can totally push you over.
i think its mostly cultural. the bsds would do the job equally as well (i used to be strongly biased in favor of bsd, but free at least has made a mess in the networking code with the addition of locks, nat, filters, firewalls, and all the other miscellany).
i'm glad you liked ipso. peter grehan and bobby minnear were two of the people who just sat down and did a solid job. i've heard less than favorable things about its more recent maintenance history.
i was one of the engineers on ipso. its not completely useless, its lovely to do network level code in, and it was about 2x faster than the freebsd it was based on (1.2) in forwarding speed. it had decent custom routing protocol implementations.
but there really isn't any need for a seperate implementation any longer. really. all you would be doing is losing out on drivers. i think its lived just as a marketing token, a random differentiator. and nokia can vaugely feel they got something from buying ipsilon. i always hear about internal struggles to replace it with linux, and remain thoroughly suprised it hasn't happened yet.
certainly there are some old a and b allocations that might be worth it, but the old/24 allocations (the swamp) are too fragmented to route globally. even getting back those/8 assignments would be difficult, there was no legal or contractual framework governing them. in fact there was a somewhat notorious incident where the ex-head-administrator of fix-west took an allocation with his name on it to a certain ip-over-cable startup and solving their addressing problem in one fell swoop.
it wouldn't have to. all that needs to be configured is a v4 tunnel endpoint address and after that you're all done. for nasty ethernet bridged networks there are all sorts of discovery options (optional dhcp fields, multicast announcments, etc).
if it were important enough and multi-hop support was a problem, one could just burn a tiny snippet of global address space, not route it in the default-free world and use it as a isp-specific service anycast address for tunnel endpoints.
this is not true. its not really fucking hard. you just have to have sufficient time and not be so lazy. it make take 4 times what you normally think, and be boring as all hell, but its totally possible.
let this be a lesson to all of you free marketeers..you know, the invisible hand. here is a whole population of lazy whiny bastards who provide almost no intrinsic value to anyone and get paid more than most.
Slashdot Mirror
this is entire estimate is meaningless. v6 was designed to
be rolled out incrementally. no one beleives that all
the endpoints are going to be upgraded.
so if some of the major backbones start peering v6, thats
a good definition for switching, but i seriously doubt its
going to involve tens of billions of dollars.
the incremental cost of new larger customers being assigned
v6 blocks instead of v4, and having to push it to the endpoints
or put in nats? the dns servers (the only thing of any
substance that was mentioned)...do the v4 roots cost billions
of dollars a year?
what, are you kidding. thats exactly what one of those godless queers
at the university would do to besmirch the name of our lord jesus
christ.
why he deserves...hey, joey, get your baseball bat...dont ask any
questions, just do it.
your statement concerning the symptom and the disease is key.
the problem isn't that we dont have enough people trying to clean
up the mess. the problem is the mess. although key distribution
is a difficult problem, the basic infrastructure needed to provide
relatively secure distributed services has existed for almost
20 years. and its still not in common use. the idea that reading
my mail can give a random person local administrative access on
my workstation is obscene at best.
hiring people to try to make things sane given the state of the
infrastructure is a losing battle...fingers in the dike. the
real question is why as a community we dont put up with the incremental
pain to put ourselves on firmer ground.
there is a reason why things tend to go as powers of two in
computer architectures. think about it, it will probably
come to you.
given that there is a general crossbar between the processors
and the memory units in the amd processors, this probably
doesn't need to be such a problem here.
but having a non power-of-two factor anywhere really screws
alot of things up. even seen a 3GB ram? a 37 bit bus?
a 27 element register file? (alright, forget this last one,
x86 is a mess)
i agree with the other posters, there must be something better
for these people to spend their time on.
you're absolutely right. by even admitting that there is
a chain of relations between species (the fossil record),
and then saying that its broken because it has holes, there
is no way that person really holds the belief that all
species were independently created.
wealth creation? i'm somewhat familiar with this capitalist fanboy term,
but in what sense does emule/edonkey create wealth? it moves bits around.
i'm sure its very nice.
some garbage collection environments let you associate
a destructor with a given object or type of object. it
still means that your connection will hang around until
some point after there are no references, but at that
point you can arrange to call close()
umm. it works the same as malloc. you get raw
pages from the operating system, and you have
to spend some of that space keeping track of
whats free.
one (overly) simple approach is to keep a list
of 'partially allocated' pages for each size
of interest (2^n). in each page there is a
freelist and free count. an allocation is thus
two indirections in the common case. a completely
empty page gets given back to the os, or put
on a list of unassigned pages. multi-page objects
are handled differently (always as a multiple of
the page size)
there are of course many other approaches. the basic
literature is very accessible if you're interested.
the solution to allocation inefficiencies due to fragmentation is well known.
it doesn't require any substantial changes to the programming model. use
a coalescing garbage collector. since you are moving objects it requires
either use of mprotect() to trap stale references or lock the object against
mutation during the copy, or the assumption that one can 'stop the world'
while copying objects.
as usual the overhead of copying can be tuned by policy, and mitigated through
the use of a standard generational implementation.
having just exposed a memory leak in ff 1.5 using svg w/ javascript, i really
do wish they would have chosen to just gc the whole thing
you're right. lightfields are cool. and they are very old news.
but its nice to actually build a compact instantaneous lightfield
capturing physical artifact, dont you think?
i worry though about the impacts on resolution. its a bit more
information than a 2d image, and the sample i saw shows it
dont forget the lowly event loop. alot of embedded
systems dont need anything like an os at all.
from my limited understanding, the back end google
clusters are designed to be fault-tolerant and
somewhat redundant.
so i expect that the model is that components
will keep dropping out until the unit is no longer
sufficiently capable or cost effective, and then
they would swap it out with a freshie.
heat is clearly an issue, which is one of the reasons
why they needed industrial designers. the most effective
solution would probably be to have smaller heat exchangers
inside the box with internal air circulation, and require
an external chilled water feed.
its a cm5. as to why anyone would choose to build that machine again,
i dont know.
are you seriously trying to claim that cray occupies a
position of marketing excellence?
from a business point of view its completely whacked. cray spends
4-5 years of time to build a machine, just to sell a very small
few of them, throw almost all of the technology away and start
over again.
and are you really trying to claim that any cray machine built in
the last 10 years has particularily good mtti rates? the sv2
really was basically unsuable for the first two years after it
was shipped. its still kind of a dubious proposition.
when is it stable?
forrest, tera is finally fully dead, which you know full well.
so its all up to you now to make it happen, you have the
company to yourself. the success or failure of which will
be entirely of your own making.
disclaimer: i worked at cray on the xt3
its not ironic at all. its a question of resources and volume.
cray has a few very bright people (still, sort of). they are
essentially a us government lab. they do a bad job, but its insane
to think that 100 people can build and maintain several different
supercomputer architectures.
a $300 opteron is almost always more effective than a $60000 X1
processor. they have alot of bright people too, and alot more of
them.
the only reason that cray still exists is support for parallelism
and the provision of high memory bandwidth systems. but even that
niche is being eroded pretty severely. the xt3 communications chip
runs at 3.5GB/s in each direction. it costs about $250 for cray to
have each of them made. for the same $250 i can buy a mellanox nic
that runs at half the speed
its no suprise that cray is using opterons. they actually got lucky
by committing to amd early and having it turn out so well.
the real question is whether there is any more room for a cray at
all. the commodity world moves so quickly. the xd machines (which
they purchased) are really their best asset, but it still hard to
justify that kind of margin for what is essentially a well
constructed cluster.
i think you're a bit off the mark. normally as you say someone
would get a cs degree, and do additional work in comp ling, then
apply to grad school with an specific bias towards that kind
of work.
however, if you really know thats what you want to do, there's alot
to be gained by putting together an interdiscplenary degree
based around your interests and getting it approved. it would
mean you would be able to take more courses in linguistics and
philosophy of language, or statistical methods, or logic, or
whatever you think you need and apply them to your degree. it means
as you suggest more work, because you probably have to develop
a basic grasp of cs as well as all those other things. but if you're
sufficiently motivated, why not? it may also help to get approval
to take grad level courses in the areas you think are necessary,
given the general paucity of undergrad level materials at that
degree of specialization.
however, if it turns out that that isn't what you end up wanting
to do forever, you have kind of painted yourself into a corner.
i also disagree about the job issue. if you're willing to accept
a narrowed field, i can think of a few companies who would love
to have comp. ling. people on staff. i've even seen job advertisements.
this person is setting a higher bar for themselves and committing to
it.
you're right in some sense, the pressure underneath the
plenum will force air through no matter what. there
are however two problems. the first is that turbulence
underneath the floor can turn the directed kinetic energy
of the air into heat...this can be a real drag. in circumstances
where you need to move alot of air, the channel may not
even be sufficiently wide.
more importantly, the air ends up coming out where the
resistance is less, leading to uneven distribution of
air. if you're grossly overbudget and just relying on
the ambient temperature of the machine room, this isn't
a problem. but when you get close to the edge it can
totally push you over.
i think its mostly cultural. the bsds would do the job
equally as well (i used to be strongly biased in favor
of bsd, but free at least has made a mess in the networking
code with the addition of locks, nat, filters, firewalls,
and all the other miscellany).
i'm glad you liked ipso. peter grehan and bobby minnear
were two of the people who just sat down and did a solid
job. i've heard less than favorable things about its more
recent maintenance history.
i was one of the engineers on ipso. its not completely useless,
its lovely to do network level code in, and it was about 2x faster
than the freebsd it was based on (1.2) in forwarding speed. it
had decent custom routing protocol implementations.
but there really isn't any need for a seperate implementation
any longer. really. all you would be doing is losing out on
drivers. i think its lived just as a marketing token, a random
differentiator. and nokia can vaugely feel they got something
from buying ipsilon. i always hear about internal struggles to
replace it with linux, and remain thoroughly suprised it hasn't
happened yet.
actually, speaking as someone who was involved in later
cray products, sgi killed the t3e.
the merger agreement with tera specifically constrained
cray from making a followon machine.
not that cray doesn't have problems....
certainly there are some old a and b allocations that might be worth it, /24 allocations (the swamp) are too fragmented to route /8 assignments would be difficult,
but the old
globally. even getting back those
there was no legal or contractual framework governing them. in fact there
was a somewhat notorious incident where the ex-head-administrator of
fix-west took an allocation with his name on it to a certain
ip-over-cable startup and solving their addressing problem in
one fell swoop.
it wouldn't have to. all that needs to be configured is a v4 tunnel
endpoint address and after that you're all done. for nasty ethernet
bridged networks there are all sorts of discovery options (optional
dhcp fields, multicast announcments, etc).
if it were important enough and multi-hop support was a problem,
one could just burn a tiny snippet of global address space, not
route it in the default-free world and use it as a isp-specific
service anycast address for tunnel endpoints.
this is not true. its not really fucking hard. you just have
to have sufficient time and not be so lazy. it
make take 4 times what you normally think, and be boring as
all hell, but its totally possible.
let this be a lesson to all of you free marketeers..you know,
the invisible hand. here is a whole population of lazy whiny
bastards who provide almost no intrinsic value to anyone and
get paid more than most.