You can't checkpoint jobs at this scale. It will take longer to checkpoint a job then to compute an answer. This is further compounded when the job takes several months to run. A 1000 node cluster is very tiny compared to the scale they're talking about.
Google is having the same problems that this article describes -- they haven't fixed it either.
If your problem domain can always be broken down into map-reduce, you can easily solve it with a hadoop-like environment to get fault tolerance. If your application falls outside of map-reduce (the applications this article is referring to), you need to start duplicating state (very expensive on systems of this scale) to recover from failures.
How do you give the work to another node when the failed node contains the only copy of its state (like in an MPI job)? Duplicating the state on multiple nodes is way too expensive.
Checkpoints will probably stick around for quite some time, but the model will need to change. Rather than serializing everything all the way down to a parallel filesystem, the data could potentially be checkpointed to a burst buffer (assuming a per-node design) or a nearby node (experimental SCR design). Of course, it's correct that even this won't scale to larger systems.
I think we'll probably have problems with getting data out to the nodes of the cluster before we start running into problems with checkpointing. The typical NFS home directory isn't going to scale. We'll need to switch over to something like udsl projections or another IO forwarding layer in the near future.
Expiration of PGP keys is a feature and does not prevent the key from being used in the future (although it should not be considered secure if used after the expiration date). The purpose is to prevent the impact of a compromised key by limiting its validity period.
Expiry can also be useful in the event that a private key is lost. Revocation of a public key requires access to the private keys.
Don't they realize this is pointless? Even if they "win", they just give more media attention to torrent sites in general. Say the pirate bay goes down (and I don't think it will) everyone will just start going to a place like mininova, or one of the other hundred popular torrent sites.
I work for the Dept of Veteran Affairs. The only change that happened (about 2 weeks ago) was the replacement of my picture id card and a hid card (rfid) with a single picture id card and rfid card combo. The only news here is that the government is making it harder for federal employees to loose their card. Now we only have one card instead of two. In addition to having an rfid card, we also have a 6-8+ digit number that we need to type into a keypad to get into more secure areas (data center, pbx room, etc). On top of all of that, you pass by about 5 or 6 cameras (that are positioned to be nearly impossible to avoid) to get anywhere near the areas in question. Finally, you'd have to have some sort of inside knowlege of the layout of the building, as all of the secured areas look less important than a dirty janitor's closet from the outside. It's not some conspiracy, it's just the government protecting people's info. I doubt any Veteran would appreciate someone stealing their personal info (as was proved recently with the laptop thing).
A stack which would implement a priority queue to hold the task structs would be pictured vertically with the last process to be put on the "top" or "back" of the queue. This would mean that the last process to run would get the highest number and the ones that are closer to being able to run get a lower one. It's not so much about tradition as it is about how the system is actually implanted and generally taught in Computer Science 101.
It makes me wonder why they haven't developed hard drives to work in a vacuum.
Hard drives work because air is there. The head basically "takes off" in a sense. It flys above the platters. In a vacuum, the head would just drag along the platter, probably destroying the drive.
the new sony dream system and the new panasonic system both have 10% of THD. but they're 600watts tho so people will still buy them.
THD, along with wattage are still only small factors in the sound a speaker puts out. materials used to make the cone, sensitivity, and resistance influence the sound enough so that your ear can actually tell the difference.
This is pretty much why bose doesn't publish the specs for their speakers. they have great marketing, but their speakers are CRAP. Most bose speakers have an eq in that completely elimanates all the sound that the speakers just can't take....bose still uses paper cone speakers and foam in their "bass modules" (they ain't subwoofers, they're only 5 & 1/4in speakers, subwoofers start at 8in). The foam dry rots after a few years and you're left with a crappy set of overpriced speakers that throw away most of the sound.
pretty much anything short of an xskey is trivial to generate a serial number for...no person in their right mind is going to do kernel level debugging to crack some app. just forcing the return address of some funct to always return some value or just nopping over a bl is much easier most of the time. if someone wanted to make another crack for your program, im sure they could do it pretty easily...except now it would only be distributed in tighter circles so you'll never know about it
Slashdot Mirror
You can't checkpoint jobs at this scale. It will take longer to checkpoint a job then to compute an answer. This is further compounded when the job takes several months to run. A 1000 node cluster is very tiny compared to the scale they're talking about.
Google is having the same problems that this article describes -- they haven't fixed it either.
If your problem domain can always be broken down into map-reduce, you can easily solve it with a hadoop-like environment to get fault tolerance. If your application falls outside of map-reduce (the applications this article is referring to), you need to start duplicating state (very expensive on systems of this scale) to recover from failures.
How do you give the work to another node when the failed node contains the only copy of its state (like in an MPI job)? Duplicating the state on multiple nodes is way too expensive.
Checkpoints will probably stick around for quite some time, but the model will need to change. Rather than serializing everything all the way down to a parallel filesystem, the data could potentially be checkpointed to a burst buffer (assuming a per-node design) or a nearby node (experimental SCR design). Of course, it's correct that even this won't scale to larger systems.
I think we'll probably have problems with getting data out to the nodes of the cluster before we start running into problems with checkpointing. The typical NFS home directory isn't going to scale. We'll need to switch over to something like udsl projections or another IO forwarding layer in the near future.
ANL's Mira is going to be roughly half as fast as LLNL's Sequoia.
If the developer is schizophrenic, then it might be n^2, otherwise it's probably (n(n-1))/2.
So, this economist and a computer scientist are sitting at a bar.... and these 5 girls walk in....
There's a reason for that edit button on top of the Wikipedia article.
I'd be happy with a TLD system based on language. Why do we need the com/net/org thing anyway. Lets just have something like
http://google.en/
http://google.it/
http://.name.language/
Actually that's the product page. The following is the actual press release:
http://www.es.northropgrumman.com/news/2008/06/144249_Northrop_Grumman-Led_Te.html
http://www.es.northropgrumman.com/solutions/hornet/index.html
Expiration of PGP keys is a feature and does not prevent the key from being used in the future (although it should not be considered secure if used after the expiration date). The purpose is to prevent the impact of a compromised key by limiting its validity period.
Expiry can also be useful in the event that a private key is lost. Revocation of a public key requires access to the private keys.
You need it for when you run G-Archiver and your emails get deleted. ;)
Don't they realize this is pointless? Even if they "win", they just give more media attention to torrent sites in general. Say the pirate bay goes down (and I don't think it will) everyone will just start going to a place like mininova, or one of the other hundred popular torrent sites.
wait, you paid $200 for a calculator?
b = $100
a = b
a^2 = ab
a^2-b^2 = ab-b^2
(a+b)(a-b) = b(a-b)
a+b = b
since a = b
b+b = b
2b = b
$200 = $100
They ripped you off. $200 is really only worth $100
I work for the Dept of Veteran Affairs. The only change that happened (about 2 weeks ago) was the replacement of my picture id card and a hid card (rfid) with a single picture id card and rfid card combo. The only news here is that the government is making it harder for federal employees to loose their card. Now we only have one card instead of two. In addition to having an rfid card, we also have a 6-8+ digit number that we need to type into a keypad to get into more secure areas (data center, pbx room, etc). On top of all of that, you pass by about 5 or 6 cameras (that are positioned to be nearly impossible to avoid) to get anywhere near the areas in question. Finally, you'd have to have some sort of inside knowlege of the layout of the building, as all of the secured areas look less important than a dirty janitor's closet from the outside. It's not some conspiracy, it's just the government protecting people's info. I doubt any Veteran would appreciate someone stealing their personal info (as was proved recently with the laptop thing).
A stack which would implement a priority queue to hold the task structs would be pictured vertically with the last process to be put on the "top" or "back" of the queue. This would mean that the last process to run would get the highest number and the ones that are closer to being able to run get a lower one. It's not so much about tradition as it is about how the system is actually implanted and generally taught in Computer Science 101.
www.tw.ioccc.org
Australia and other Pacific
www.au.ioccc.org
Europe
www.de.ioccc.org
www.es.ioccc.org
www.gr.ioccc.org
North America
www0.us.ioccc.org
www1.us.ioccc.org
um, take a look at MDoom (search on google). It's a clone of doom for your celly telly.
Hard drives work because air is there. The head basically "takes off" in a sense. It flys above the platters. In a vacuum, the head would just drag along the platter, probably destroying the drive.
the scary part is, I've done that in my FREE time and enjoyed it...so :P
When I think oil in the radiator, I think blown head gasket. It could of been a serious problem.
1% is if you're lucky
the new sony dream system and the new panasonic system both have 10% of THD. but they're 600watts tho so people will still buy them.
THD, along with wattage are still only small factors in the sound a speaker puts out. materials used to make the cone, sensitivity, and resistance influence the sound enough so that your ear can actually tell the difference.
This is pretty much why bose doesn't publish the specs for their speakers. they have great marketing, but their speakers are CRAP. Most bose speakers have an eq in that completely elimanates all the sound that the speakers just can't take....bose still uses paper cone speakers and foam in their "bass modules" (they ain't subwoofers, they're only 5 & 1/4in speakers, subwoofers start at 8in). The foam dry rots after a few years and you're left with a crappy set of overpriced speakers that throw away most of the sound.
pretty much anything short of an xskey is trivial to generate a serial number for...no person in their right mind is going to do kernel level debugging to crack some app. just forcing the return address of some funct to always return some value or just nopping over a bl is much easier most of the time. if someone wanted to make another crack for your program, im sure they could do it pretty easily...except now it would only be distributed in tighter circles so you'll never know about it
s/hacker/cracker/
or
s/cracker/hacker/
it just doesn't matter anymore...the media has blurred the definition