As many people have pointed out Beowulf clusters cannot be restricted, precisely because they are clusters and not one machine.
Whether it is practical to prevent the export of Beowulf systems, and whether the government will attempt to prohibit the export of Beowulf systems are two completely separate questions.
Consider the longstanding battles over encryption software export. Does anyone honestly believe forreign governments couldn't manage to get their hands on the 128-bit version of Netscape? Come on! Yet, the government still interfered here. It's important to remember this.
He said that they cost less, offer full compatability, and offer better performance than Intel.
It turned out only one of these things were true. It did indeed cost less.
In 1996 I bought a 100MHz 686 which Cyrix branded a PR-133. I ran with it as my primary system for two years and never had *any* problems with hardware/software compatibility. Prior to that I had purchased Cyrix brand 586 and 486 systems. The 486 ran fine. I think there was some funny business going on with the 586, though, since I had to underclock the thing to get it to compile a kernel. I suspect the vendor, though, not Cyrix. With none of these three chips did I ever experience a software compatibility problem.
Yes, I know that some companies have said that their product doesn't work with the chips, but I've also had many companies tell me that their products don't work with Linux, while in fact they worked just fine. If a company hasn't tested something, they frequently claim it to be incompatible, since, that's the safe call to make. For some reason, companies seem to have a very hard time saying "We don't know.".
The Cyrix 686 chips *do* perform significantly better than Intel pentium chips running at the same clock *for certain types of calculations*.
If you aren't happy with the fact that salesman treated you as a "mark", that's understandable. But when presented with marketing information by a salesman, you have to be aware that his motivating factor is to sell a product. If you would prefer a salesman that more honestly represents his product, then by all means, shop somewhere else, but don't fault the original manufacturer.
Ok, so, assume apple releases a freely redistributable binary only version of a QT player for Linux...
It's going to run on my UltraSparc isn't it?
How about that Corel Netwinder with the StrongArm in it. It'll work there, too, right?
And on the Alpha?
PPC? That's Apple's home turf, where Linux can go head to head with MacOS. Apple will release a Linux PPC version, too?
Companies don't want to support Linux with commercial applications. It's just too expensive to maintain. Sure, some companies will cater to the largest Linux demographic, Linux-i386, but, since when is Linux about only supporting the largest demographic?
You can get a 36 GB IDE drive from most drive companies now and rumor has it IBM has 50 and 70 GB IDE drives now (I don't know if that's really true, I'll have to check it out.) How much hard disk storage do you actually need? Seems to me you could slap 2 or 3 50+ GB drives in your system and be set for the foreseeable future
A few times 50 GB might be fine for you at home, but if you want to set up a server at an institution which is supposed to handle 100+ people who run scientific computing applications which routinely produce multi-gigabytes worth of information, then this is nowhere near enough.
You always need more room.
One of the experiments I work on has been recording about 30 GB of data every day for the past 3.5 years. We keep it on tape, of course. There's no way we can afford to keep it all on disk, but we would if we could.
You always need more room.
In 1988 a friend of my Dad's came by and seeing our new PC told us that there was just no way we would ever fill its hard disk. We just could never produce that much data. The drive was nearly as large as DOS 3.2 would allow for a single partition. If it was any larger we would have had to partition it into smaller pieces which DOS could handle. It was a 30 MB disk. Needless to say, we eventually filled it.
The 486 SX was in reality a 486 DX with a faulty math co-processor/floating point unit. Intel had no plans to make SX/DX models until the first crop of 486's had faulty fpu's.
What are you smoking? The designation SX/DX has nothing to do with a math coprocessor.
"SX" stands for "Single eXternal" and "DX" for "Double eXternal".
With 286 CPU's, both the motherboard and CPU had 16 bit busses, but, with the introduction of the 386 processor, the CPU bus doubled in width to 32 bits. The 386 was manufactured in two varieties, SX and DX. The DX version was all 32 bits everywhere. The SX version used a 32 bit internal bus on the CPU, but sat on a 16 bit motherboard. The SX CPU allowed 386 systems to run on cheaper motherboards and lowered the bar for the 386 entry to the low end market.
Neither the 386 SX or DX have an onboard FPU. If you want a hardware FPU with a 386 system, you need to add a 387 regardless of the type of 386 you have.
The 486 systems also came in SX and DX varieties to allow the 486 to run on cheaper motherboards as well. The 486 DX was the first x86 to include an onboard FPU. This increased the price of the chip quite a bit, but, since since this was targeted exclusively at the high end market, Intel felt this was acceptable.
The SX is not just a DX without a math-coprocessor. You need to add lots of extra circuitry to properly mate the 16 bit motherboard bus with the 32 bit CPU bus. These circuits don't exist on the DX chips.
You're getting mixed up between kilobytes per second and kilobits per second.
No, you're getting mixed up.
The statement I made is that the drive is 4.6GB not 4.6 Gb like the title says. GB is gigabytes and Gb is gigabits. I'm not mixing them up, the poster is.
4.6Gb? There's no way you can fit 80+ hours of music in only 4.6Gb. 80 hours is 288000 seconds, so 4.6Gb would only be about 16Kbps. Have you ever heard a 16Kbps mp3? Terrible.
The drive in this thing is 4.6GB not 4.6Gb. That's what the article says, and that's the only thing that makes any sense.
Huh? 15k???? Maybe for a rackmount unit or something... but PC DVD-RAM drives start at bout $370.00...
DVD RAM and DVD ROM are different technologies. Not all DVD players can read DVD RAM. (In fact, I believe most cannot.) Not to mention the fact that the drive which you're referring to can only write at a density of 5.2 GB per disk. (2.6 GB/side). The $15K model is needed for writing the standard 18 GB DVD-ROM disks.
There isn't much need to say more. You want fast code? You have to listen to music that has some tempo too it. Anything less than 160 bpm is too slow to code too.
I now have a different graphics card, and I'm back on XF86 (those vmware accellerations are hard to beat), but I wouldn't hesitate to go to a commercial X server if I started having problems again. Nor should you.
That's a bit of a religious issue. If you believe in free software and the philosophies behind it, that's a very strong reason not to go commercial.
If I were having trouble running my graphics card under X-Free, and I had $150 with which to rectify the situation, I would have no trouble deciding what to do. Buy a new graphics card with better XFree support.
I don't use commercial software when there's a viable free alternative. Nor should you.
I think, from what I've read, the best buy for general use today are those cards based on the Trident 4D Wave NX chipset.
They're relatively inexpensive, have fancy features like digital output, are well supported under Linux, and are produced by a company which has not only released full technical documentation for these cards (without any silly NDA's), but fully GPL'd Linux drivers have been written and released by Trident themselves. You can't ask for much more than that.
Ha ha! Do you really think that RPM/DEB is the best thing for 600MB game??
Yes.
At least you can't have both packages on one CD.
Yes, you could..deb files (I assume.rpm, too) can access external data files via scripts. The bulk of the data for a 600MB game is just that, data. It's usually architecture and OS independent at that. This large data set could be in any format on the cd, and manipulated by the comparatively small.deb/.rpm files.
This is currently how Debian distributes stuff it can't distribute. (sort of) Like Real Audio, for example.
Er.. Yes we do. Tried to install xyz deb on Redhat? vice versa? alien is not a panacea. How about either on Slackware? Redhat RPMS on SuSE
Let me rephrase my post. The point of a distribution is to package and distribute. Debian does an excellent job of providing Debian compatible installers for lots of commercial apps. (Netscape, Real Audio, etc.) By going through Debian, I know that these installers aren't going to fsck up my system because my system wasn't configured the way the vendor was expecting. I know it's not going to overwrite any files that might be under another package system control, and I know I'm going to be able to automatically get updates with "apt-get update" without needing to track down ten million different sites all over the net and see if a new version has been released or not.
tar.gz is the only distribution neutral system - and for what do you need package management when commercial software is supposed to install into/opt/packagname anyway?
For dependency checking and for easy updating.
Its also a hassle to package for each obscure little variant of Linux with different packagmanagement, libc's, et al, and then try and do installation support.
Of course, it's a hassle for the vendor, but not for the distributor. That's the distributor's job! Let Debian make debs and RH the rpms, etc. etc.
What newbie ever learned to use their package manager at levels lower than point and click?
This is a terrible idea. Loki should release their software in standard.deb/.rpm package formats and let the distributions handle installation, dependencies, removal, and upgrades. This is what distributions do! Let them do it! We don't need an "Install Shield".
WE ALREADY HAVE INSTALLERS which are best suited to the distributions on which they run!
You want this to make it easier for the newbie? Why ask the newbie to learn yet another installer when he already knows one for his distribution?
Although this is good news, I don't know if it really warrents the term "breakthrough".
Do the math. The standard battery got 3:04, or 184 minutes. The new battery got 15:01, or 901 minutes. What really matters though, is not the difference in run time, but the difference in run time per unit mass. The standard battery was only 0.6625 lbs, so a battery made from hooking 2.2 lbs worth of standard batteries together should be expected to run for 184*2.2/0.6625 = 611 minutes, or 10:11, compared to 15:01. So this is a 47% improvement.
Not bad, but I don't know how they can justify the statement that they have "more than twice the energy density" just by looking at these numbers.
Of course, these devices need to prove themselves on many other fronts before they become practical. Cost, long term reliability, charging times, and malfunction possibilities all need to be considered before endorsing a technology such as this.
Imagine trying to support linux over the phone....what fricking nightmare.
Phone support is difficult under the best of circumstances, but, IMNSHO, linux phone support is *WAY* easier than Windows.
For one thing, "type this:" is much easier to express clearly over the phone than "find an icon that looks like a computer and is labeled 'System', then click on the tab labeled 'Device Manager', then look for a picture of a..."
But the real big win for Linux is that, if the network is up, a remote techincal support person can actually log in there and see for himself, directly what the problem is. Linux just dominates Windows when it comes to remote administration.
If I were a large PC vendor and wanted to ship supported Linux desktops to newbies, I'd install an admin account on each machine which would allow tech support to log in and identify problems. This would allow tech support not only to fix problems, but also to do periodic maintainence, sanity checks, and upgrades according to a service contract selected by the customer. (Customers would, of course, be able to disable such an account at any time.)
A small boot partition on the hard disk could be reserved to boot the system with a standard configuration which would guarentee that the system would retain enough functionality to dial into the tech support network regardless of the abuses the user applies to the operating system.
Think of the ads: Having problems? Just type "fixme" at the lilo prompt and our expert technical support staff will be connected to your machine in minutes!
I think this sort of thing will eventually become the norm, once people get used to the idea. Hackers will always take care of their own machines, but the masses will let someone else do the dirty work. I don't see this happening with Windows anytime soon, though.
Ever install a second drive in a Windows machine and have the CD-Rom change from D to E, causing half a million short-cuts to break? I'm glad I don't have to deal with that stuff much anymore.
this isn't HTML pages that this guy is linking to. He is linking directly to the Quicktime files.
So what? It should be ok to link to text/html, but not quicktime files?
The point is that Universal set up a web server and they set it up in such a way that any browser that requests any document on that server would be served that document, be it HTML, Quicktime or otherwise. If they then get upset that the web server is doing just that, then who do they have to blame.
All linking is is telling the browser where it can get a particular piece of information.
The important thing to remember here is who is in control. Universal. They run the server. They control the content and behavior of the server. The government uses law to protect people from things beyond their control. There is no excuse for Universal whining to the courts to change the fact that their servers will spit out information to whoever wants it.
In fact, I would expect that sites that host copies of content, and serve it on their own bandwidth dollars, would actually be more objectionable to the original content providers, since this would remove their control over the information. Universal can always 404 these links any time they like, but they can't do that if someone is hosting copies.
This is obviously a case of the PHB's not understanding the technical details.
The Trident 4DWave NX based cards support this. They're still a little hard to find, but you can order them from Hoontech at http://www.hoontech.com. They offer quite a bang for the buck.
Trident actually wrote a GPL'd Linux driver for this thing, so it's a good card to buy.
Slashdot Mirror
As many people have pointed out Beowulf clusters cannot be restricted, precisely because they are clusters and not one machine.
Whether it is practical to prevent the export of Beowulf systems, and whether the government will attempt to prohibit the export of Beowulf systems are two completely separate questions.
Consider the longstanding battles over encryption software export. Does anyone honestly believe forreign governments couldn't manage to get their hands on the 128-bit version of Netscape? Come on! Yet, the government still interfered here. It's important to remember this.
He said that they cost less, offer full compatability, and offer better performance than Intel.
It turned out only one of these things were true. It did indeed cost less.
In 1996 I bought a 100MHz 686 which Cyrix branded a PR-133. I ran with it as my primary system for two years and never had *any* problems with hardware/software compatibility. Prior to that I had purchased Cyrix brand 586 and 486 systems. The 486 ran fine. I think there was some funny business going on with the 586, though, since I had to underclock the thing to get it to compile a kernel. I suspect the vendor, though, not Cyrix. With none of these three chips did I ever experience a software compatibility problem.
Yes, I know that some companies have said that their product doesn't work with the chips, but I've also had many companies tell me that their products don't work with Linux, while in fact they worked just fine. If a company hasn't tested something, they frequently claim it to be incompatible, since, that's the safe call to make. For some reason, companies seem to have a very hard time saying "We don't know.".
The Cyrix 686 chips *do* perform significantly better than Intel pentium chips running at the same clock *for certain types of calculations*.
If you aren't happy with the fact that salesman treated you as a "mark", that's understandable. But when presented with marketing information by a salesman, you have to be aware that his motivating factor is to sell a product. If you would prefer a salesman that more honestly represents his product, then by all means, shop somewhere else, but don't fault the original manufacturer.
I have three different versions of Netscape running on RedHat 5.2, 6.0, and 6.1, and they *all* die during Java loads.
Must be a RH problem. Works fine from all of my Debian boxes running NS 4.5-4.7.
Ok, so, assume apple releases a freely redistributable binary only version of a QT player for Linux...
It's going to run on my UltraSparc isn't it?
How about that Corel Netwinder with the StrongArm in it. It'll work there, too, right?
And on the Alpha?
PPC? That's Apple's home turf, where Linux can go head to head with MacOS. Apple will release a Linux PPC version, too?
Companies don't want to support Linux with commercial applications. It's just too expensive to maintain. Sure, some companies will cater to the largest Linux demographic, Linux-i386, but, since when is Linux about only supporting the largest demographic?
You can get a 36 GB IDE drive from most drive companies now and rumor has it IBM has 50 and 70 GB IDE drives now (I don't know if that's really true, I'll have to check it out.) How much hard disk storage do you actually need? Seems to me you could slap 2 or 3 50+ GB drives in your system and be set for the foreseeable future
A few times 50 GB might be fine for you at home, but if you want to set up a server at an institution which is supposed to handle 100+ people who run scientific computing applications which routinely produce multi-gigabytes worth of information, then this is nowhere near enough.
You always need more room.
One of the experiments I work on has been recording about 30 GB of data every day for the past 3.5 years. We keep it on tape, of course. There's no way we can afford to keep it all on disk, but we would if we could.
You always need more room.
In 1988 a friend of my Dad's came by and seeing our new PC told us that there was just no way we would ever fill its hard disk. We just could never produce that much data. The drive was nearly as large as DOS 3.2 would allow for a single partition. If it was any larger we would have had to partition it into smaller pieces which DOS could handle. It was a 30 MB disk. Needless to say, we eventually filled it.
You always need more room.
I'm pretty sure the 386 (i386) didn't *HAVE* the SX/DX markings until much later in the production run (say, around the time that 486's got them).
Wrong. Just check the CPU Info Center listings. The 386 SX was introduced in 1988, while the 486 didn't show up until three years later in 1991.
The 486 SX was in reality a 486 DX with a faulty math co-processor/floating point unit. Intel had no plans to make SX/DX models until the first crop of 486's had faulty fpu's.
What are you smoking? The designation SX/DX has nothing to do with a math coprocessor.
"SX" stands for "Single eXternal" and "DX" for "Double eXternal".
With 286 CPU's, both the motherboard and CPU had 16 bit busses, but, with the introduction of the 386 processor, the CPU bus doubled in width to 32 bits. The 386 was manufactured in two varieties, SX and DX. The DX version was all 32 bits everywhere. The SX version used a 32 bit internal bus on the CPU, but sat on a 16 bit motherboard. The SX CPU allowed 386 systems to run on cheaper motherboards and lowered the bar for the 386 entry to the low end market.
Neither the 386 SX or DX have an onboard FPU. If you want a hardware FPU with a 386 system, you need to add a 387 regardless of the type of 386 you have.
The 486 systems also came in SX and DX varieties to allow the 486 to run on cheaper motherboards as well. The 486 DX was the first x86 to include an onboard FPU. This increased the price of the chip quite a bit, but, since since this was targeted exclusively at the high end market, Intel felt this was acceptable.
The SX is not just a DX without a math-coprocessor. You need to add lots of extra circuitry to properly mate the 16 bit motherboard bus with the 32 bit CPU bus. These circuits don't exist on the DX chips.
You're getting mixed up between kilobytes per second and kilobits per second.
No, you're getting mixed up.
The statement I made is that the drive is 4.6GB not 4.6 Gb like the title says. GB is gigabytes and Gb is gigabits. I'm not mixing them up, the poster is.
4.6Gb? There's no way you can fit 80+ hours of music in only 4.6Gb. 80 hours is 288000 seconds, so 4.6Gb would only be about 16Kbps. Have you ever heard a 16Kbps mp3? Terrible.
The drive in this thing is 4.6GB not 4.6Gb. That's what the article says, and that's the only thing that makes any sense.
Huh? 15k???? Maybe for a rackmount unit or
something... but PC DVD-RAM drives start at
bout $370.00...
DVD RAM and DVD ROM are different technologies. Not all DVD players can read DVD RAM. (In fact, I believe most cannot.) Not to mention the fact that the drive which you're referring to can only write at a density of 5.2 GB per disk. (2.6 GB/side). The $15K model is needed for writing the standard 18 GB DVD-ROM disks.
http://www.hornet.org
There isn't much need to say more. You want fast code? You have to listen to music that has some tempo too it. Anything less than 160 bpm is too slow to code too.
I now have a different graphics card, and I'm back on XF86 (those vmware accellerations are hard to beat), but I wouldn't hesitate to go to a commercial X server if I started having problems again. Nor should you.
That's a bit of a religious issue. If you believe in free software and the philosophies behind it, that's a very strong reason not to go commercial.
If I were having trouble running my graphics card under X-Free, and I had $150 with which to rectify the situation, I would have no trouble deciding what to do. Buy a new graphics card with better XFree support.
I don't use commercial software when there's a viable free alternative. Nor should you.
I think, from what I've read, the best buy for general use today are those cards based on the Trident 4D Wave NX chipset.
They're relatively inexpensive, have fancy features like digital output, are well supported under Linux, and are produced by a company which has not only released full technical documentation for these cards (without any silly NDA's), but fully GPL'd Linux drivers have been written and released by Trident themselves. You can't ask for much more than that.
Check out those made by Hoontech for example.
See John Fulmer's lengthy review.
Ha ha! Do you really think that RPM/DEB is the best thing for 600MB game??
.deb files (I assume .rpm, too) can access external data files via scripts. The bulk of the data for a 600MB game is just that, data. It's usually architecture and OS independent at that. This large data set could be in any format on the cd, and manipulated by the comparatively small .deb/.rpm files.
Yes.
At least you can't have both packages on one CD.
Yes, you could.
This is currently how Debian distributes stuff it can't distribute. (sort of) Like Real Audio, for example.
Er.. Yes we do. Tried to install xyz deb on Redhat? vice versa? alien is not a panacea. How about either on Slackware? Redhat RPMS on SuSE
/opt/packagname anyway?
Let me rephrase my post. The point of a distribution is to package and distribute. Debian does an excellent job of providing Debian compatible installers for lots of commercial apps. (Netscape, Real Audio, etc.) By going through Debian, I know that these installers aren't going to fsck up my system because my system wasn't configured the way the vendor was expecting. I know it's not going to overwrite any files that might be under another package system control, and I know I'm going to be able to automatically get updates with "apt-get update" without needing to track down ten million different sites all over the net and see if a new version has been released or not.
tar.gz is the only distribution neutral system - and for what do you need package management when commercial software is supposed to install into
For dependency checking and for easy updating.
Its also a hassle to package for each obscure little variant of Linux with different packagmanagement, libc's, et al, and then try and do installation support.
Of course, it's a hassle for the vendor, but not for the distributor. That's the distributor's job! Let Debian make debs and RH the rpms, etc. etc.
What newbie ever learned to use their package manager at levels lower than point and click?
apt-get update
apt-get install package
apt-get dist-upgrade
That's it. That's all you need to know for basic Debian package management. How is that hard for a newbie?
This is a terrible idea. Loki should release their software in standard .deb/.rpm package formats and let the distributions handle installation, dependencies, removal, and upgrades. This is what distributions do! Let them do it! We don't need an "Install Shield".
WE ALREADY HAVE INSTALLERS which are best suited to the distributions on which they run!
You want this to make it easier for the newbie? Why ask the newbie to learn yet another installer when he already knows one for his distribution?
Exactly how is this new, and what is it about this that qualifies this man as a "genious"?
I've seen better looking graphics in 5 year old euro-demos running on a 386 DOS box.
Although this is good news, I don't know if it really warrents the term "breakthrough".
Do the math. The standard battery got 3:04, or 184 minutes. The new battery got 15:01, or 901 minutes. What really matters though, is not the difference in run time, but the difference in run time per unit mass. The standard battery was only 0.6625 lbs, so a battery made from hooking 2.2 lbs worth of standard batteries together should be expected to run for 184*2.2/0.6625 = 611 minutes, or 10:11, compared to 15:01. So this is a 47% improvement.
Not bad, but I don't know how they can justify the statement that they have "more than twice the energy density" just by looking at these numbers.
Of course, these devices need to prove themselves on many other fronts before they become practical. Cost, long term reliability, charging times, and malfunction possibilities all need to be considered before endorsing a technology such as this.
Imagine trying to support linux over the phone....what fricking nightmare.
Phone support is difficult under the best of circumstances, but, IMNSHO, linux phone support is *WAY* easier than Windows.
For one thing, "type this:" is much easier to express clearly over the phone than "find an icon that looks like a computer and is labeled 'System', then click on the tab labeled 'Device Manager', then look for a picture of a..."
But the real big win for Linux is that, if the network is up, a remote techincal support person can actually log in there and see for himself, directly what the problem is. Linux just dominates Windows when it comes to remote administration.
If I were a large PC vendor and wanted to ship supported Linux desktops to newbies, I'd install an admin account on each machine which would allow tech support to log in and identify problems. This would allow tech support not only to fix problems, but also to do periodic maintainence, sanity checks, and upgrades according to a service contract selected by the customer. (Customers would, of course, be able to disable such an account at any time.)
A small boot partition on the hard disk could be reserved to boot the system with a standard configuration which would guarentee that the system would retain enough functionality to dial into the tech support network regardless of the abuses the user applies to the operating system.
Think of the ads: Having problems? Just type "fixme" at the lilo prompt and our expert technical support staff will be connected to your machine in minutes!
I think this sort of thing will eventually become the norm, once people get used to the idea. Hackers will always take care of their own machines, but the masses will let someone else do the dirty work. I don't see this happening with Windows anytime soon, though.
Ever install a second drive in a Windows machine and have the CD-Rom change from D to E, causing half a million short-cuts to break? I'm glad I don't have to deal with that stuff much anymore.
this isn't HTML pages that this guy is linking to. He is linking directly to the Quicktime files.
So what? It should be ok to link to text/html, but not quicktime files?
The point is that Universal set up a web server and they set it up in such a way that any browser that requests any document on that server would be served that document, be it HTML, Quicktime or otherwise. If they then get upset that the web server is doing just that, then who do they have to blame.
All linking is is telling the browser where it can get a particular piece of information.
The important thing to remember here is who is in control. Universal. They run the server. They control the content and behavior of the server. The government uses law to protect people from things beyond their control. There is no excuse for Universal whining to the courts to change the fact that their servers will spit out information to whoever wants it.
In fact, I would expect that sites that host copies of content, and serve it on their own bandwidth dollars, would actually be more objectionable to the original content providers, since this would remove their control over the information. Universal can always 404 these links any time they like, but they can't do that if someone is hosting copies.
This is obviously a case of the PHB's not understanding the technical details.
Hasn't this been available for a long time?
BitWizard has had a patch on their page for a long time and it supports drives from several different manufacturers, too!
What is new about Nathan's patch that isn't provided by the BitWizard patch?
Silly me, I always thought the bottle neck was the user...
The Trident 4DWave NX based cards support this. They're still a little hard to find, but you can order them from Hoontech at http://www.hoontech.com. They offer quite a bang for the buck.
Trident actually wrote a GPL'd Linux driver for this thing, so it's a good card to buy.
Sure, it's only 1.0E-7 molecules per mole, but it is _something_
Actually, it's 1.0E-7 moles per liter. That's about 1.0E+15 molecules per mole