actually its quite the converse. a decentralized distribution system will in the limit always be more efficient than sending N copies across the accesslink from the same server.
if everyone can find a copy close to me in the topology, then overall we can save something up to copies*diameter of load. and that load will be more evenly spread out.
a single server is actually the least efficient way of distributing information to a large number of clients possible.
not even considering benefits such as tolerance of server and network failures, anonymity, etc
the lights on the cm-5 were just driven by one of the suns inside the cabinet. they were totally gratuitous.there were a few different driver programs you could select.
on the cm-2, they were actually driven by the processors themselves, and you could tell what application was running by looking at the pattern....but they too were effectively gratuitous. i do remember writing a tetris on the cm-2, but there wasn't much space.
you didn't live through networking 'revolutions' like ATM. at that time we were told repeatedly that IP may have been a nice playground, but the telcos, who really know how to run a service, were going to take over now. IP and TCP were never going to be suitable for large scale business deployment.
'at&t moves voice service to ip' would have been a hilarious gag article 10 years ago, now no one even blinks.
i've worked in various capacities for contractors of the dod (primarily darpa), for my entire technical life (> 15 years).
only because there is no other place to do interesting research and advanced development. there are plenty of positive things that can be done with my work, but no one else has the money to allow me to pursue it.
i heard a rumor that they have one or more CMs. i actually know how to maintain them, does anyone know if they are looking to restore one? it would be fun.
Re:Why not just....
on
fvwm Turns Ten
·
· Score: 0, Redundant
marginally useful? what else could you possibly need. you can move, resize, iconify, and uniconfy windows
actually no. i was in the group that did the kernel work for ipso. it has a custom ip forwarding path and forwarding table machinery. the routing is done using a largely rewritten version of...gated
these three things and the management system make ipso a good software routing platform.
which doesn't really offset the cost of what is a pretty sluggish pc
i work at cray, on the red storm machine that is replacing a 40Tflop machine build it 1996. that asci red machine is now something like 8th in the us..and as others have said, commodity clusters have a hard time applying those resources to solving a problem scalably, the communications latency is just too high
Slashdot Mirror
catamount is a slightly rewarmed cougar
tflops was a dsm os called osf/ad, its being replaced by suse with some minimal cluster extensions
actually its quite the converse. a decentralized distribution system will in the limit always be more efficient than sending N copies across the accesslink from the same server.
if everyone can find a copy close to me in the topology, then overall we can save something up to copies*diameter of load. and that load will be more evenly spread out.
a single server is actually the least efficient way of distributing information to a large number of clients possible.
not even considering benefits such as tolerance of server and network failures, anonymity, etc
the lights on the cm-5 were just driven by one of the suns inside the cabinet. they were totally gratuitous.there were a few different driver programs you could select.
on the cm-2, they were actually driven by the processors themselves, and you could tell what application was running by looking at the pattern....but they too were effectively gratuitous. i do remember writing a tetris on the cm-2, but there wasn't much space.
you didn't live through networking 'revolutions' like ATM. at that time we were told repeatedly that IP may have been a nice playground, but the telcos, who really know how to run a service, were going to take over now. IP and TCP were never going to be suitable for large scale business deployment.
'at&t moves voice service to ip' would have been a hilarious gag article 10 years ago, now no one even blinks.
i've worked in various capacities for contractors of the dod (primarily darpa), for my entire technical life (> 15 years).
only because there is no other place to do interesting research and advanced development. there are plenty of positive things that can be done with my work, but no one else has the money to allow me to pursue it.
try continued fractions
thats nothing. they were originally going to call
it 'red hammer'
i heard a rumor that they have one or more CMs. i actually know how to maintain them, does anyone know if they are looking to restore one? it would be fun.
marginally useful? what else could you possibly
need. you can move, resize, iconify, and uniconfy
windows
-proud twm user for at least 15 years (i think)
actually no. i was in the group that did the kernel work for ipso. it has a custom ip forwarding path and forwarding table machinery. the routing is done using a largely rewritten version of...gated
these three things and the management system make ipso a good software routing platform.
which doesn't really offset the cost of what is a pretty sluggish pc
there will never be a keyboard so lovely, although a kinesis almost makes do
i work at cray, on the red storm machine that is ..and as others have said, commodity clusters have a hard time applying those resources to solving a problem scalably, the communications latency is just
replacing a 40Tflop machine build it 1996. that
asci red machine is now something like 8th in the
us
too high
this is a GPL project being done by a small company, Clustered File Systems....www.lustre.org
this is not a 'normal' filesystem. it stripes data across distributed disk servers and will provide clustered metadata services in the future