In theory, yes. However programmers usually do stupid mistakes, like assuming that sizeof (int) == sizeof (void *) and so they think they can cast pointers to int and the other way around, while on a typical LP64 platform (like AMD64), ints are 32 bits wide and pointers 64 bits, so the cast will not work as expected.
All in all, nothing new here, well written portable code just need a recompilation, and everything else will need to be debugged.
It is important in scenarios where a lot of small burst transfers
happen almost all the time, like on a busy file server.
In this case switching from 150 MB/s sata links to 300 MB/s ones will
reduce the PCI/PCI-X/PCI-e bus utilization. Here is a realistic example:
each 160 ms, a block of 8 MB of data needs to be sent to a sata disk,
that means on average 6.25 blocks are sent per second (1000 / 160).
If a 150 MB/s sata link is used, a practical 120+ MB/s burst data transfer
rate can be achieved, which means the data block would be transferred
in about 67 ms (8 / 120 * 1000). Since 6.25 blocks need to be sent per
second, a total of 419 ms (67 * 6.25) is spent doing the transfer from
RAM, over the PCI/PCI-X/PCI-e bus, over the sata link, and to the disk.
The bus utilization is about 41.9 %.
If a 300 MB/s sata link is used, a practical 240+ MB/s burst data transfer
rate can be achieved, which means the data block would be transferred
in about 33 ms (8 / 240 * 1000). Since 6.25 blocks need to be sent per
second, a total of 206 ms (33 * 6.25) is spent doing the transfer from
RAM, over the PCI/PCI-X/PCI-e bus, over the sata link, and to the disk.
The bus utilization is about 20.6 %.
In this example, using 300 MB/s sata links reduced the PCI/PCI-X/PCI-e
bus utilization from 41.9 % to 20.6 %. Which is important in case other
devices need to use it (e.g. network cards in a file server, etc).
Please note that in the case of the slowest bus (PCI 32-bit 33 MHz), this
example doesn't make sense because a 150 MB/s sata link (120 MB/s practical
data throughput) already fully saturates the bus. The example remains valid for any other case (PCI 66 MHz, or PCI-X, or PCI-e).
<<
No, it's not important. The on-disk cache isn't there for throughput,
it's there because of the disk's rotational latency.
>>
I am not speaking about sustained throughput, I am speaking about burst
transfers. So in this regard, I maintain that yes, a 300 MB/s sata link
will make a difference and will improve (some) workloads.
<<
the OS doesn't know the disk geometry and so *can't* do the type of caching
the on-disk cache does.
>>
Modern OSes already know how to optimize the disks' heads movement.
They can do it since LBA48 sector addresses are ordered from the inner
cylinders to the outer cylinders. This allow Linux for example (I am familiar
with its implementation) to use its block layer to reorder I/O requests
in order to minimize disks' heads movement. In fact some kernel hackers
even argue that NCQ won't bring such a major improvement because this
block layer optimization is already very effective. You could even further
argue that the OS can only do a better job than the on-disk cache, because
it has a higher level view of what is happenning on the system: applications
requiring better throughput, or better latency, etc, and can by consequent
better tune the behavior of its I/O scheduler.
Every single review of storage technologies I have read in the past 2 years state at least one (IMHO very important) incorrect fact. This is no exception with this one:
Page 2: "The fastest 3.5" SATA drives do not exceed 85 MB/s. A data transfer rate of 300 MB/s between a PC and a SATA drive cannot thus be matched by the speed of a SATA drive." Yes it can. When data is exchanged to/from the disk's cache, data throughput of 250+ MB/s can be achieved for a fraction of a second. Even if it's only for a fraction of a second, it is still important (else manufacturers would not even put cache memory on disks).
Lucent (to MS): Microsoft, you violated our MPEG2 patent. We demand you recall all 360s. Now. Microsoft: What !? We are afraid it's not going to be possible. Lucent (angrily): Do it now ! Or we will sue you ! Sony (pointing his finger toward MS): LOLLLL Huhuhuhu Huhu ! Lucent (to Sony): You too ! Sony: Bastards...
I suggest we all start using "snarfed" as a synonym for "slashdotted". As in "we have figuratively stolen your server('s bandwith)" == "we have slashdotted your server" == "we have snarfed your server". I like this word better anyway. Thank you, OP, for your contribution to the/. subculture language !
PS: Don't worry about your server, it will be back up online soon when the story will leave the front page.
Pleaaase Slashdot editors, on my knees I beg you to remove the pink color. It hurts our eyes reaaally hard. April Fool's Day is supposed to be a joke, not a torture. Please be clement.
I would like to re-post what I posted the 1st time (this article is a dupe):
This is an interesting coincidence because I used to reflect deeply on this exact subject a few years ago: what if a supercomputer could simulate a human ? I'll be honest here: I am literally _astounded_ to discover that this scientific team has successfuly simulated a virus. I didn't thought supercomputers were powerful enough for such a task. I just finished reading some articles about the experience and I now understand why this has been possible: they used some empirical functions instead of implementing exact physical laws (would have required much more computing power) and they also simulated the virus for only 50 billionths of a second. But still, they seem to have successfuly simulated life.
Most people don't realize the significance of this event, it means that given enough computing resources we could theoretically simulate humans ! One day we will have enough computing power to run such a simulation. And when it will be done, this human life simulation will have the potential to prove (or disprove) that humans are "just" a bunch of atoms following physical laws and nothing more.
This is huge. Think about it. I know this may sound sad, but personally I am convinced that any life form, including humans, is just that a complex assembly of atoms following physical laws, there is no soul, no afterlife, etc. This human life supercomputer will prove I am right:)
I have always wondered what those guys suing for anything _really_ think ? For example, does this guy honestly thought Google was violating his copyright ? Or did he sue just to give a try and maybe obtain easy money via financial compensation ?
No. The pb with tapes is that their data throughput is slow, they are not random access devices, not as reliable as what you would think, more expensive than hard drives, and currently losing popularity. I interviewed about 2 year ago at a petroleum company which has Petabytes of data to backup weekly and they were precisely migrating to hard drive-based backup systems for these very reasons. The higher rate of failure of hard drives can easily be compensated by making 2 or 3 copies of your backups. The funny thing is that hard drives are so cheap that such a solution would even be cheaper than a tap-based system.
Plus the advantage of hard drive-based backup systems, or online storage as the industry call it, is that your data is immediately available: no need to manually insert a tape in a reader, etc. Hard disks contructors have even created a new line of products to accomodate the nead of backup systems, google for "nearline storage".
Current hard disks sell for 40c/gig. If you plan to keep your data for >3 months, it is more economical to use you harddisks. Though your solution has other advantages (data accessible from anywhere, no need to administer yourself your servers, etc).
This is an interesting coincidence because I used to reflect deeply on this exact subject a few years ago: what if a supercomputer could simulate a human ? I'll be honest here: I am literally _astounded_ to discover that this scientific team has successfuly simulated a virus. I didn't thought supercomputers were powerful enough for such a task. I just finished reading some articles about the experience and I now understand why this has been possible: they used some empirical functions instead of implementing exact physical laws (would have required much more computing power) and they also simulated the virus for only 50 billionths of a second. But still, they seem to have successfuly simulated life.
Most people don't realize the significance of this event, it means that given enough computing resources we could theoretically simulate humans ! One day we will have enough computing power to run such a simulation. And when it will be done, this human life simulation will have the potential to prove (or disprove) that humans are "just" a bunch of atoms following physical laws and nothing more.
This is huge. Think about it. I know this may sound sad, but personally I am convinced that any life form, including humans, is just that a complex assembly of atoms following physical laws, there is no soul, no afterlife, etc. This human life supercomputer will prove I am right:)
See my comment about 4 PCIe cards here. The whole point of what I am proposing is precisely to be able to use regular commodity hardware to do tasks that, nowadays, can only be accomplished using high-end expensive gear. This is critical for some businesses. See Google ? Their whole architecture is built with commodity hardware.
Of course I know that 12.8 GB/s is a theoretical value. But even reaching the third of that value is totally impossible with current mobos:( Yet I could build a box with 4.2 GB/s of potential I/O: four 16-port PCIe SATA cards with 16*4 = 64 disks. A modern SATA disk can sustain about 65 MB/s of read/write operations. And 64 * 65 MB/s = 4.2 GB/s. Such boxes do exist today, but cannot realize their full potential because of slow PCI-X busses (PCI-X 2 can alleviate the situation, but PCI-X 2 mobo are VERY RARE).
I would love to see a quad-Opteron mobo with four x16 PCIe slots but arranged in a way that traffic is spread across all HT links. So that I could use it to put 4 PCIe SATA cards, and have the highest possible read/write I/O throughput for a Linux software RAID array. Hardware RAID is out of the question, since no constructor offers a way to create arrays of disks across 3 or more cards. An Opteron has 3 HT links, 2 of them could be used as coherent links to other CPU's, and 1 of them could be used as a link to an external PCIe bridge chipset. The solution I would like to see implemented is one where 4 PCIe bridge chipsets would be connected to their own Opteron, via their own HT link. And each PCIe bridge chipset could provide at least one 16x slot.
Some numbers: each of the four x16 PCIe bus would allow for 2500 MT/s * 16 bits / 8 = 5000 MB/s of traffic in each direction. And each of the 4 HT links: 1600 MT/s * 16 bits / 8 = 3200 MB/s. The global amount of I/O would be 3200 MB/s * 4 = 12.8 GB/s in each direction ! (HT links are the bottleneck). To resolve this bottleneck AMD would either need to increase their width from 16x16 to 32x32 bits or need to increase the signal freq from 800 MHz to 1.25 GHz (current limit is 1 GHz for coherent links and 800 MHz for the ones facing outside worlds -- chipsets seem to lag a little bit regarding HT frequency).
But for some reason no constructor has ever designed such a board (Tyan only did it with 2 PCIe chipsets on their S2895 mobo). Why oh why is that the case ?! Seems like nobody understands the true potential of HT. This could provide a low-cost solution to so many perf issues I have seen in the various companies I have worked for... Argh !
There is a grammatical error, the correct writing is: L'iPod est mort, vive l'iPod.
See ? Being French is advantageous. Anytime someone tries to write something in french on/. you can be sure to find an error. So just do like me:
1- Reply to fix the error.
2- Wait for the nice "+5, Informative" mod.
3- ???
4- Karma increased !
Slashdot Mirror
In theory, yes. However programmers usually do stupid mistakes, like assuming that sizeof (int) == sizeof (void *) and so they think they can cast pointers to int and the other way around, while on a typical LP64 platform (like AMD64), ints are 32 bits wide and pointers 64 bits, so the cast will not work as expected.
All in all, nothing new here, well written portable code just need a recompilation, and everything else will need to be debugged.
It is important in scenarios where a lot of small burst transfers happen almost all the time, like on a busy file server. In this case switching from 150 MB/s sata links to 300 MB/s ones will reduce the PCI/PCI-X/PCI-e bus utilization. Here is a realistic example: each 160 ms, a block of 8 MB of data needs to be sent to a sata disk, that means on average 6.25 blocks are sent per second (1000 / 160).
In this example, using 300 MB/s sata links reduced the PCI/PCI-X/PCI-e bus utilization from 41.9 % to 20.6 %. Which is important in case other devices need to use it (e.g. network cards in a file server, etc).
Please note that in the case of the slowest bus (PCI 32-bit 33 MHz), this example doesn't make sense because a 150 MB/s sata link (120 MB/s practical data throughput) already fully saturates the bus. The example remains valid for any other case (PCI 66 MHz, or PCI-X, or PCI-e).
I am not speaking about sustained throughput, I am speaking about burst transfers. So in this regard, I maintain that yes, a 300 MB/s sata link will make a difference and will improve (some) workloads.
Modern OSes already know how to optimize the disks' heads movement. They can do it since LBA48 sector addresses are ordered from the inner cylinders to the outer cylinders. This allow Linux for example (I am familiar with its implementation) to use its block layer to reorder I/O requests in order to minimize disks' heads movement. In fact some kernel hackers even argue that NCQ won't bring such a major improvement because this block layer optimization is already very effective. You could even further argue that the OS can only do a better job than the on-disk cache, because it has a higher level view of what is happenning on the system: applications requiring better throughput, or better latency, etc, and can by consequent better tune the behavior of its I/O scheduler.
Every single review of storage technologies I have read in the past 2 years state at least one (IMHO very important) incorrect fact. This is no exception with this one:
Page 2: "The fastest 3.5" SATA drives do not exceed 85 MB/s. A data transfer rate of 300 MB/s between a PC and a SATA drive cannot thus be matched by the speed of a SATA drive." Yes it can. When data is exchanged to/from the disk's cache, data throughput of 250+ MB/s can be achieved for a fraction of a second. Even if it's only for a fraction of a second, it is still important (else manufacturers would not even put cache memory on disks).
Ooh yeah ! I saw this porn mov... oh you mean silicon as in silicon chips ?
Nevermind.
Lucent (to MS): Microsoft, you violated our MPEG2 patent. We demand you recall all 360s. Now.
Microsoft: What !? We are afraid it's not going to be possible.
Lucent (angrily): Do it now ! Or we will sue you !
Sony (pointing his finger toward MS): LOLLLL Huhuhuhu Huhu !
Lucent (to Sony): You too !
Sony: Bastards...
I suggest we all start using "snarfed" as a synonym for "slashdotted". As in "we have figuratively stolen your server('s bandwith)" == "we have slashdotted your server" == "we have snarfed your server". I like this word better anyway. Thank you, OP, for your contribution to the /. subculture language !
PS: Don't worry about your server, it will be back up online soon when the story will leave the front page.
Or they were demo'ing Intel processors.
And this process was named... "LINUX" !
I can't wait for Britannica's reply to Nature's reply about Britannica's criticism of the Nature Britannica-Wikipedia comparison !
Hahaha :) Blondes are going to click-and-click-and-click... ad infinitum without ever realizing it :)
Chained-slashdotting:
n 1: cf. above slashdot article.
Pleaaase Slashdot editors, on my knees I beg you to remove the pink color. It hurts our eyes reaaally hard. April Fool's Day is supposed to be a joke, not a torture. Please be clement.
This is an interesting coincidence because I used to reflect deeply on this exact subject a few years ago: what if a supercomputer could simulate a human ? I'll be honest here: I am literally _astounded_ to discover that this scientific team has successfuly simulated a virus. I didn't thought supercomputers were powerful enough for such a task. I just finished reading some articles about the experience and I now understand why this has been possible: they used some empirical functions instead of implementing exact physical laws (would have required much more computing power) and they also simulated the virus for only 50 billionths of a second. But still, they seem to have successfuly simulated life.
Most people don't realize the significance of this event, it means that given enough computing resources we could theoretically simulate humans ! One day we will have enough computing power to run such a simulation. And when it will be done, this human life simulation will have the potential to prove (or disprove) that humans are "just" a bunch of atoms following physical laws and nothing more.
This is huge. Think about it. I know this may sound sad, but personally I am convinced that any life form, including humans, is just that a complex assembly of atoms following physical laws, there is no soul, no afterlife, etc. This human life supercomputer will prove I am right :)
Do you have something bigger (for my porn collection) ? Thanks !
I have always wondered what those guys suing for anything _really_ think ? For example, does this guy honestly thought Google was violating his copyright ? Or did he sue just to give a try and maybe obtain easy money via financial compensation ?
What do you mean by "is supposed to" ? Oh... you mean "has to". What OS are you using again ?
No. The pb with tapes is that their data throughput is slow, they are not random access devices, not as reliable as what you would think, more expensive than hard drives, and currently losing popularity. I interviewed about 2 year ago at a petroleum company which has Petabytes of data to backup weekly and they were precisely migrating to hard drive-based backup systems for these very reasons. The higher rate of failure of hard drives can easily be compensated by making 2 or 3 copies of your backups. The funny thing is that hard drives are so cheap that such a solution would even be cheaper than a tap-based system.
Plus the advantage of hard drive-based backup systems, or online storage as the industry call it, is that your data is immediately available: no need to manually insert a tape in a reader, etc. Hard disks contructors have even created a new line of products to accomodate the nead of backup systems, google for "nearline storage".
Current hard disks sell for 40c/gig. If you plan to keep your data for >3 months, it is more economical to use you harddisks. Though your solution has other advantages (data accessible from anywhere, no need to administer yourself your servers, etc).
This is an interesting coincidence because I used to reflect deeply on this exact subject a few years ago: what if a supercomputer could simulate a human ? I'll be honest here: I am literally _astounded_ to discover that this scientific team has successfuly simulated a virus. I didn't thought supercomputers were powerful enough for such a task. I just finished reading some articles about the experience and I now understand why this has been possible: they used some empirical functions instead of implementing exact physical laws (would have required much more computing power) and they also simulated the virus for only 50 billionths of a second. But still, they seem to have successfuly simulated life.
Most people don't realize the significance of this event, it means that given enough computing resources we could theoretically simulate humans ! One day we will have enough computing power to run such a simulation. And when it will be done, this human life simulation will have the potential to prove (or disprove) that humans are "just" a bunch of atoms following physical laws and nothing more.
This is huge. Think about it. I know this may sound sad, but personally I am convinced that any life form, including humans, is just that a complex assembly of atoms following physical laws, there is no soul, no afterlife, etc. This human life supercomputer will prove I am right :)
I confirm :)
See my comment about 4 PCIe cards here. The whole point of what I am proposing is precisely to be able to use regular commodity hardware to do tasks that, nowadays, can only be accomplished using high-end expensive gear. This is critical for some businesses. See Google ? Their whole architecture is built with commodity hardware.
Of course I know that 12.8 GB/s is a theoretical value. But even reaching the third of that value is totally impossible with current mobos :( Yet I could build a box with 4.2 GB/s of potential I/O: four 16-port PCIe SATA cards with 16*4 = 64 disks. A modern SATA disk can sustain about 65 MB/s of read/write operations. And 64 * 65 MB/s = 4.2 GB/s. Such boxes do exist today, but cannot realize their full potential because of slow PCI-X busses (PCI-X 2 can alleviate the situation, but PCI-X 2 mobo are VERY RARE).
I would love to see a quad-Opteron mobo with four x16 PCIe slots but arranged in a way that traffic is spread across all HT links. So that I could use it to put 4 PCIe SATA cards, and have the highest possible read/write I/O throughput for a Linux software RAID array. Hardware RAID is out of the question, since no constructor offers a way to create arrays of disks across 3 or more cards. An Opteron has 3 HT links, 2 of them could be used as coherent links to other CPU's, and 1 of them could be used as a link to an external PCIe bridge chipset. The solution I would like to see implemented is one where 4 PCIe bridge chipsets would be connected to their own Opteron, via their own HT link. And each PCIe bridge chipset could provide at least one 16x slot.
Some numbers: each of the four x16 PCIe bus would allow for 2500 MT/s * 16 bits / 8 = 5000 MB/s of traffic in each direction. And each of the 4 HT links: 1600 MT/s * 16 bits / 8 = 3200 MB/s. The global amount of I/O would be 3200 MB/s * 4 = 12.8 GB/s in each direction ! (HT links are the bottleneck). To resolve this bottleneck AMD would either need to increase their width from 16x16 to 32x32 bits or need to increase the signal freq from 800 MHz to 1.25 GHz (current limit is 1 GHz for coherent links and 800 MHz for the ones facing outside worlds -- chipsets seem to lag a little bit regarding HT frequency).
But for some reason no constructor has ever designed such a board (Tyan only did it with 2 PCIe chipsets on their S2895 mobo). Why oh why is that the case ?! Seems like nobody understands the true potential of HT. This could provide a low-cost solution to so many perf issues I have seen in the various companies I have worked for... Argh !
There is a grammatical error, the correct writing is: L'iPod est mort, vive l'iPod.
See ? Being French is advantageous. Anytime someone tries to write something in french on /. you can be sure to find an error. So just do like me:
1- Reply to fix the error.
2- Wait for the nice "+5, Informative" mod.
3- ???
4- Karma increased !