"Couldn't the OS instruct the drive to perform those speculative reads? If anything, the OS should have more information about whether those next blocks will be needed than the drive itself."
Even if the OS knew better than the hard drive (sometimes it does, sometimes it doesn't), it would require the CPU to think about it and formulate the requests, and it would eat up bandwidth transfering the stuff to the host (including memory bandwidth).
As I said, the hard drive is often in a position to get the gains for free or close to free, while the OS has to contend with the limited speed of the bus and all sorts of other limitations.
"I wonder if there are any drives sold in both 2MB and 8MB cache versions, and how their benchmarks compare."
I have two very similar drives (Seagate 80 GB) where the only differences are 8 MB cache instead of 2 and SATA instead of ATA-100. Obviously there's no way for someone like me to measure how much SATA contributes to the performance, but that drive kicks the crap out of the ATA one even when it's used with an OS that doesn't support command queuing. The extra bandwidth alone wouldn't make that much of a difference (as evidenced by the small increase in performance from ATA-66 to 100 and 100 to 133).
Sometimes it's a BIOS issue. I have a P2 that won't detect an 80 GB drive, but once the OS is booted it works fine. Obivously, this means I have to boot with another drive or set the "limited capacity" jumper (which limits you to 32 GB).
Sometimes it's an ATA thing. Without 48-bit addressing, the drives are limited to 137 GB. I understand this requires hardware support. I am confused about your Pentium 120 being able to do 200 GB (which implies 48-bit addressing).
For example, speculative reads that are essentially free when the heads are in position, but would be much more expensive after the drive rotates out of position.
Just look at the same benchmarks on otherwise identical drives with different amounts of cache.
Lack of loyalty to employers is due to lack of loyalty to employees. I'd be loyal to my employer if I didn't see my coworkers getting canned because we're a few percent overbudget, even when it leaves us unable to live up some pretty sweet contracts that we already have.
SSH tunnel to a proxy server. Unless they have spyware on my computer (our software is unstable enough as it is), there's no way to know what I'm doing.
I think a beta means "enough of the stuff is there for us to want the public to test it". RC means "if the problems are cleared up we can release this unchanged".
"Your ATA controller points are invalid, as they're also problems that your PII must deal with as well. If not, it had to be a pro level board that happened to get BIOS updates by the manufacturer THAT insanely long after the production time."
You're partly correct; the motherboard does not recognize drives larger than 32 gb and therefore cannot boot from them. If that were the end of it I'd be out of luck. However, there's two ways around this that I can think of:
1. Use a drive smaller than 32 gb to boot the OS, then let the OS deal with it. In my case, this allows ATA drives up to 137 gigs* at ATA-33 speed. For SATA drives with a controller card, this allows the drive to operate at full SATA speed up to whatever the max size is with SATA. You still can't get past 137 gb per drive without a hardware upgrade.
2. Let the BIOS on the SATA controller card deal with booting the OS from the SATA drive. The motherboard must be set to allow an external BIOS to handle the boot procedure, one usually does this by telling it there's a SCSI card to boot from. I guess an SATA card acts the same. This is what I do.
"Firewire networking ha sbeen around for ages on both Mac's and PC's, NIC problem solved."
Unfortunately, not all the computers I wish to put on the network have FireWire (in fact the only one that does is the iBook, it still isn't standard in the PC world). While adding Firewire to my own computers is essentially the same as adding gigabit ethernet to the server, it doesn't help other people's laptops and computers. I can't reasonably expect them to have a laptop with FireWire that's correctly configured to use such a network.
"Next up we have the harddrive issue, external enclosure, if it's a firewall, it doesn't need to move, also,"
That could work.
"what does a firtewall need a 160gig drive foor? Or a super fast one for that mater. I have a G3 tower with ATA 33 and have no issues with useing it as a firewall, with firewire networking."
It's a fileserver as well.
"Why do you need USB 2.0 with firewire built it? I see no advantages to the USB 2.0 Hi Speed specification over Firewire."
Advantage 1: USB keys. A gig of stuff takes a loooong time to transfer at USB 1.1 speeds, and even if they made equivilant FireWire drives, which they don't, I'd get a USB one because it would work on more computers.
Advantage 2: USB printers. USB 2.0 would make this significantly faster. While I haven't set this up yet, it was part of my motivation to upgrade to USB 2.0
"SATA is a waste IMO, SCSI is a much better choice, but this is the only thing you mentioned that holds water. ATA is a shity choice for a primary hard drive as only 1 drive can talk at a time on an ATA bus, but I believe Apple puts it's optical drives on a seperate bus."
That's a very good point. However, it has nothing to do with SATA, which is immune to any problems resulting from drives sharing a channel due to the fact that it supports a maximum of one drive per channel.
As for SCSI... it costs a lot more. Enough that it's not even a consideration for me.
"PowerMacs give you PCI slots. My G3 (the same age as your iMac) now has Firewire, USB 2, and upgraded Ethernet. The machine is still going, handling tasks I offload from my powerbook (long term file storage/backup, music, long file transfers, etc)"
As I said, PowerMacs cost a lot extra for someone just looking for an upgradable computer. That premium is quite large when all you're looking for is a computer that can be upgraded to the next version of USB in 3-4 years.
"Im wondering what you are serving from this machine."
gah, I don't even remember it off the top of my head. It's primarily a web and NFS server, but as far as I can remember I run these servers (most are not externally accessible):
DHCP, FTP, DNS, SSH, WWW (with dynamic content), PostgreSQL (for the dynamic content), CVS, NFS, NTP (so timestamps for all the NFS mounts agree), FTP proxy for the firewall, Squid proxy that I SSH tunnel to, and the obvious firewall/NAT.
"Headaches?"
Not after I installed OpenBSD.
"Im sure that a P2 400 cant handle Gigbit while writing to a SATA drives/card. while handling traffic from 2 nics. BOTTLENECKS???? i think would be the term here."
I get 20 mbyte/s on NFS mounts on my workstation. As the theoretical max with 100 megabit is half that, I'd say it helped. Since that's pretty close to the best NFS can do, I'm happy with it.
The cable modem is 512 kbit/s up and 5 mbit/s down, the traffic generated by it when it's going flat out isn't significant compared to the internal LAN and the NFS traffic.
"It sounds suspisciously like you don't want a consumer machine... If you wanted all this, would you buy a low level $600 Dell machine *then* upgrade everything? If you want pro-level equipment, you probably want to *buy* pro-level equipment"
What's with the labels?
Apple likes to label the iMac a "consumer" machine and the PowerMac a "pro" machine because they get better margins on the "pro" machine. They artificially enforce this by leaving out features that they can add for free (dual monitor support on iBooks) or for very little cost (gigabit ethernet support) on the "consumer" machine, forcing anyone that requires any one of these features to pay for all of them.
No, this distinction is entirely an arbitrary one. The Apple "pro" machines don't have ECC memory or several other features expected of a workstation. They're merely at the high end of consumer machines.
If Dell or some other OEM sold a machine that was close to what I wanted, such that I could get what I wanted cheaper than building it myself or buying a higher end machine, what's wrong with that?
If Dell is selling a machine for $600 that has everything I want but gigabit ethernet, I say $625 for a machine that has everything I want is a much better deal than $1500.
"Very true. However, that hasn't ever presented a problem to me. By the time the "spiffy port" has changed, I'm buying a new computer anyways. But as always, YMMV."
The machine I described has been going strong for a very long time now, and it does the job as well as anything I could put together new. There's no reason to replace it.
"I can see your point, but realize, your strategy is not Apple's strategy. Because people who want fancy things buy the fancy computers, Apple is able to make a profit. Imagine if their $499 Mac was as expandable as their PowerMac- it would cannabalize their PowerMac sales."
Yup. But the obvious question is whether or not other OEMs are cannabalizing their PowerMac sales.
"In the end, if you don't like it- you don't have to buy it. What works best for you, be it a Mac or PC or NeXT Cube or whatever, is what you should use."
Yes.
This is the point that Mac zealots don't like. It's the point that zealots in general don't like. I use OpenBSD, Linux, and MacOS because each platform is good at one job and bad at the other two.
"If you find that portage is an acceptable substitute for debian's package management, then there you go."
As a former Gentoo user, I think that's pretty unlikely. It's a good tool, but the portage tree suffers from a chronic lack of testing. I had problems that would have been caught if a package had been tested before it was released (eg the version of fam that kde 3.2 wanted was masked after kde 3.2 was released).
The problem with non-Apple memory is that it's your problem when it breaks. One of the times I had my logic board replaced, they broke my memory. I was out of pocket for the replacement, but if it had been Apple memory I would have just taken it back again.
"
Just out of curiousity, what are you going to use those PCI slots for ?
There is already NIC, Firewire, USB, Sound and Video cards onboard. I've had several macs, and i've never installed a single addon card in any of them.
The only thing i've ever come up with was to use one as a firewall, in that case a second NIC would be desirable, but otherwise?"
Uhg... I hate it when people say that.
To most people, PCI slots don't matter. To a minority, they do matter, and to that minority, the lack may prevent them from buying a Mac. In my case, I have a tendency to upgrade older machines and move them into a server role as I replace them on the desktop, and this is not possible with the inexpensive Macs being discussed in this article.
For example, consider my current firewall/server machine and the upgrades I have done, relative to an iMac from the same time period. It's a Pentium 2 400 mhz from 1997 or so.
-Add another NIC so I can use it as a firewall... impossible on the iMac. -Add an SATA card... impossible on the iMac. -Add a 160 gb hard drive... impossible on the iMac as the ATA controllers of the time could not handle drives bigger than 128 gb. -Use the drive at full speed... impossible on the iMac because the ATA controllers of the time were limited to ATA-33. -Now using 2 hard drives... impossible on the iMac. -Upgrade the second NIC to gigabit... impossible on the iMac. Impossible on current iMacs too. -Upgrade the USB to USB 2.0... impossible on the iMac.
It's not that Apple computers don't have all the spiffy ports, it's that they can't be upgraded later on when the definition of "spiffy port" changes.
People usually argue that enthusiasts like myself should be buying PowerMacs, but the whole point is that a $500 PC is just as capable of doing these things as a $2000 PowerMac. PowerMacs have many benefits, but you pay for a lot of benefits that you don't need to buy the one benefit that you do.
"My main worry is that I'll switch, drink the Apple kool-aid, then wake up one morning and think "The novelty of the pretty eye-candy has worn off now. What do I have that I would not have with GNU/Linux with (say) KDE?""
I drank the kool-aid, and the novelty doesn't really wear off as with 10.3. I think MacOS is preferable to KDE if only by a little bit.
The problems are the fiddly little things. They're sluggish coming out with the new Java versions. A lot of software doesn't work quite right (GIMP didn't have a good port for a long time, OpenOffice is a death trap).
As a Debian user, you will miss the package management.
I've been a Mac user for almost 2 years now, and while I like the machine, I don't like it any more than my Linux machine. As an experienced Linux user, it's not that much harder for me to use, easier in some ways (I'm also a Debian user), and a whole lot more versatile. I couldn't honestly do without either of them. I also probably can't do without my OpenBSD machine, which I probably like more than both of them.
I guess I just don't think any one OS can do the trick.
I am unfamiliar with the Windows implementation details, but on UNIX the mprotect(2) system call is used, which allows a userspace program to modify the permissions on a given page.
Apparently it doesn't break much on UNIX because many platforms are quite lax about caching policies on executable code. If the OS doesn't flush the cache, code may break, and mprotect is used to take an opportunity to flush the cache.
Or something like that. I forget the details of what Theo said in the talk.
It's not "akin". OpenBSD's W^X is policy. AMD's NX bit is a processor feature that makes policy like that easier to implement. Microsoft's policy is different (and not as strong, only the stack is protected), but uses the same processor features.
Slashdot Mirror
"Couldn't the OS instruct the drive to perform those speculative reads? If anything, the OS should have more information about whether those next blocks will be needed than the drive itself."
Even if the OS knew better than the hard drive (sometimes it does, sometimes it doesn't), it would require the CPU to think about it and formulate the requests, and it would eat up bandwidth transfering the stuff to the host (including memory bandwidth).
As I said, the hard drive is often in a position to get the gains for free or close to free, while the OS has to contend with the limited speed of the bus and all sorts of other limitations.
"I wonder if there are any drives sold in both 2MB and 8MB cache versions, and how their benchmarks compare."
I have two very similar drives (Seagate 80 GB) where the only differences are 8 MB cache instead of 2 and SATA instead of ATA-100. Obviously there's no way for someone like me to measure how much SATA contributes to the performance, but that drive kicks the crap out of the ATA one even when it's used with an OS that doesn't support command queuing. The extra bandwidth alone wouldn't make that much of a difference (as evidenced by the small increase in performance from ATA-66 to 100 and 100 to 133).
Sometimes it's a BIOS issue. I have a P2 that won't detect an 80 GB drive, but once the OS is booted it works fine. Obivously, this means I have to boot with another drive or set the "limited capacity" jumper (which limits you to 32 GB).
Sometimes it's an ATA thing. Without 48-bit addressing, the drives are limited to 137 GB. I understand this requires hardware support. I am confused about your Pentium 120 being able to do 200 GB (which implies 48-bit addressing).
"Is the on drive cache really that useful?"
Yes.
For example, speculative reads that are essentially free when the heads are in position, but would be much more expensive after the drive rotates out of position.
Just look at the same benchmarks on otherwise identical drives with different amounts of cache.
I do everything through an SSH tunnel to a proxy server. And I'm not even conspiring.
Keeps the noise level up.
Portable Firefox, PuTTY, ssh tunnel to a proxy server. :)
Unless they have spyware on the computer, they can't tell what I'm doing. And our software is unstable enough as it is.
Lack of loyalty to employers is due to lack of loyalty to employees. I'd be loyal to my employer if I didn't see my coworkers getting canned because we're a few percent overbudget, even when it leaves us unable to live up some pretty sweet contracts that we already have.
SSH tunnel to a proxy server. Unless they have spyware on my computer (our software is unstable enough as it is), there's no way to know what I'm doing.
I think a beta means "enough of the stuff is there for us to want the public to test it". RC means "if the problems are cleared up we can release this unchanged".
In Calgary, we consider anything above 25 to be too hot.
I know nothing about the climate of that part of Germany, but there's a good chance it's comfortable for the Germans.
The editors generally pick it up.
:)
Or rather, they might pick it up but there's nothing anyone else can do about it but whine.
"Your ATA controller points are invalid, as they're also problems that your PII must deal with as well. If not, it had to be a pro level board that happened to get BIOS updates by the manufacturer THAT insanely long after the production time."
You're partly correct; the motherboard does not recognize drives larger than 32 gb and therefore cannot boot from them. If that were the end of it I'd be out of luck. However, there's two ways around this that I can think of:
1. Use a drive smaller than 32 gb to boot the OS, then let the OS deal with it. In my case, this allows ATA drives up to 137 gigs* at ATA-33 speed. For SATA drives with a controller card, this allows the drive to operate at full SATA speed up to whatever the max size is with SATA. You still can't get past 137 gb per drive without a hardware upgrade.
2. Let the BIOS on the SATA controller card deal with booting the OS from the SATA drive. The motherboard must be set to allow an external BIOS to handle the boot procedure, one usually does this by telling it there's a SCSI card to boot from. I guess an SATA card acts the same. This is what I do.
Neither case requires a BIOS update.
* - I mistakenly said 128 gb in an above post.
"Firewire networking ha sbeen around for ages on both Mac's and PC's, NIC problem solved."
Unfortunately, not all the computers I wish to put on the network have FireWire (in fact the only one that does is the iBook, it still isn't standard in the PC world). While adding Firewire to my own computers is essentially the same as adding gigabit ethernet to the server, it doesn't help other people's laptops and computers. I can't reasonably expect them to have a laptop with FireWire that's correctly configured to use such a network.
"Next up we have the harddrive issue, external enclosure, if it's a firewall, it doesn't need to move, also,"
That could work.
"what does a firtewall need a 160gig drive foor? Or a super fast one for that mater. I have a G3 tower with ATA 33 and have no issues with useing it as a firewall, with firewire networking."
It's a fileserver as well.
"Why do you need USB 2.0 with firewire built it? I see no advantages to the USB 2.0 Hi Speed specification over Firewire."
Advantage 1: USB keys. A gig of stuff takes a loooong time to transfer at USB 1.1 speeds, and even if they made equivilant FireWire drives, which they don't, I'd get a USB one because it would work on more computers.
Advantage 2: USB printers. USB 2.0 would make this significantly faster. While I haven't set this up yet, it was part of my motivation to upgrade to USB 2.0
"SATA is a waste IMO, SCSI is a much better choice, but this is the only thing you mentioned that holds water. ATA is a shity choice for a primary hard drive as only 1 drive can talk at a time on an ATA bus, but I believe Apple puts it's optical drives on a seperate bus."
That's a very good point. However, it has nothing to do with SATA, which is immune to any problems resulting from drives sharing a channel due to the fact that it supports a maximum of one drive per channel.
As for SCSI... it costs a lot more. Enough that it's not even a consideration for me.
"PowerMacs give you PCI slots. My G3 (the same age as your iMac) now has Firewire, USB 2, and upgraded Ethernet. The machine is still going, handling tasks I offload from my powerbook (long term file storage/backup, music, long file transfers, etc)"
As I said, PowerMacs cost a lot extra for someone just looking for an upgradable computer. That premium is quite large when all you're looking for is a computer that can be upgraded to the next version of USB in 3-4 years.
"Im wondering what you are serving from this machine."
gah, I don't even remember it off the top of my head. It's primarily a web and NFS server, but as far as I can remember I run these servers (most are not externally accessible):
DHCP, FTP, DNS, SSH, WWW (with dynamic content), PostgreSQL (for the dynamic content), CVS, NFS, NTP (so timestamps for all the NFS mounts agree), FTP proxy for the firewall, Squid proxy that I SSH tunnel to, and the obvious firewall/NAT.
"Headaches?"
Not after I installed OpenBSD.
"Im sure that a P2 400 cant handle Gigbit while writing to a SATA drives/card. while handling traffic from 2 nics. BOTTLENECKS???? i think would be the term here."
I get 20 mbyte/s on NFS mounts on my workstation. As the theoretical max with 100 megabit is half that, I'd say it helped. Since that's pretty close to the best NFS can do, I'm happy with it.
The cable modem is 512 kbit/s up and 5 mbit/s down, the traffic generated by it when it's going flat out isn't significant compared to the internal LAN and the NFS traffic.
"It sounds suspisciously like you don't want a consumer machine... If you wanted all this, would you buy a low level $600 Dell machine *then* upgrade everything? If you want pro-level equipment, you probably want to *buy* pro-level equipment"
What's with the labels?
Apple likes to label the iMac a "consumer" machine and the PowerMac a "pro" machine because they get better margins on the "pro" machine. They artificially enforce this by leaving out features that they can add for free (dual monitor support on iBooks) or for very little cost (gigabit ethernet support) on the "consumer" machine, forcing anyone that requires any one of these features to pay for all of them.
No, this distinction is entirely an arbitrary one. The Apple "pro" machines don't have ECC memory or several other features expected of a workstation. They're merely at the high end of consumer machines.
If Dell or some other OEM sold a machine that was close to what I wanted, such that I could get what I wanted cheaper than building it myself or buying a higher end machine, what's wrong with that?
If Dell is selling a machine for $600 that has everything I want but gigabit ethernet, I say $625 for a machine that has everything I want is a much better deal than $1500.
"Very true. However, that hasn't ever presented a problem to me. By the time the "spiffy port" has changed, I'm buying a new computer anyways. But as always, YMMV."
The machine I described has been going strong for a very long time now, and it does the job as well as anything I could put together new. There's no reason to replace it.
"I can see your point, but realize, your strategy is not Apple's strategy. Because people who want fancy things buy the fancy computers, Apple is able to make a profit. Imagine if their $499 Mac was as expandable as their PowerMac- it would cannabalize their PowerMac sales."
Yup. But the obvious question is whether or not other OEMs are cannabalizing their PowerMac sales.
"In the end, if you don't like it- you don't have to buy it. What works best for you, be it a Mac or PC or NeXT Cube or whatever, is what you should use."
Yes.
This is the point that Mac zealots don't like. It's the point that zealots in general don't like. I use OpenBSD, Linux, and MacOS because each platform is good at one job and bad at the other two.
"If you find that portage is an acceptable substitute for debian's package management, then there you go."
As a former Gentoo user, I think that's pretty unlikely. It's a good tool, but the portage tree suffers from a chronic lack of testing. I had problems that would have been caught if a package had been tested before it was released (eg the version of fam that kde 3.2 wanted was masked after kde 3.2 was released).
OS X will fly on the G4 processor they're talking about (1.25 ghz) if you give it at least 512 mb of memory.
The problem with non-Apple memory is that it's your problem when it breaks. One of the times I had my logic board replaced, they broke my memory. I was out of pocket for the replacement, but if it had been Apple memory I would have just taken it back again.
The Apple people broke my 3rd party memory upgrade when they were fixing my logic board.
To most people, PCI slots don't matter. To a minority, they do matter, and to that minority, the lack may prevent them from buying a Mac. In my case, I have a tendency to upgrade older machines and move them into a server role as I replace them on the desktop, and this is not possible with the inexpensive Macs being discussed in this article.
For example, consider my current firewall/server machine and the upgrades I have done, relative to an iMac from the same time period. It's a Pentium 2 400 mhz from 1997 or so.
-Add another NIC so I can use it as a firewall... impossible on the iMac.
-Add an SATA card... impossible on the iMac.
-Add a 160 gb hard drive... impossible on the iMac as the ATA controllers of the time could not handle drives bigger than 128 gb.
-Use the drive at full speed... impossible on the iMac because the ATA controllers of the time were limited to ATA-33.
-Now using 2 hard drives... impossible on the iMac.
-Upgrade the second NIC to gigabit... impossible on the iMac. Impossible on current iMacs too.
-Upgrade the USB to USB 2.0... impossible on the iMac.
It's not that Apple computers don't have all the spiffy ports, it's that they can't be upgraded later on when the definition of "spiffy port" changes.
People usually argue that enthusiasts like myself should be buying PowerMacs, but the whole point is that a $500 PC is just as capable of doing these things as a $2000 PowerMac. PowerMacs have many benefits, but you pay for a lot of benefits that you don't need to buy the one benefit that you do.
"My main worry is that I'll switch, drink the Apple kool-aid, then wake up one morning and think "The novelty of the pretty eye-candy has worn off now. What do I have that I would not have with GNU/Linux with (say) KDE?""
I drank the kool-aid, and the novelty doesn't really wear off as with 10.3. I think MacOS is preferable to KDE if only by a little bit.
The problems are the fiddly little things. They're sluggish coming out with the new Java versions. A lot of software doesn't work quite right (GIMP didn't have a good port for a long time, OpenOffice is a death trap).
As a Debian user, you will miss the package management.
I've been a Mac user for almost 2 years now, and while I like the machine, I don't like it any more than my Linux machine. As an experienced Linux user, it's not that much harder for me to use, easier in some ways (I'm also a Debian user), and a whole lot more versatile. I couldn't honestly do without either of them. I also probably can't do without my OpenBSD machine, which I probably like more than both of them.
I guess I just don't think any one OS can do the trick.
I am unfamiliar with the Windows implementation details, but on UNIX the mprotect(2) system call is used, which allows a userspace program to modify the permissions on a given page.
Apparently it doesn't break much on UNIX because many platforms are quite lax about caching policies on executable code. If the OS doesn't flush the cache, code may break, and mprotect is used to take an opportunity to flush the cache.
Or something like that. I forget the details of what Theo said in the talk.
It's not "akin". OpenBSD's W^X is policy. AMD's NX bit is a processor feature that makes policy like that easier to implement. Microsoft's policy is different (and not as strong, only the stack is protected), but uses the same processor features.
They would last years (assuming the dust doesn't screw up the wheels), be able to travel much farther, and opperate at night.
Now I know that DragonFlyBSD is truly one of the BSD pantheon.
It has zealotous trolls.