Vista Rant

[GAHD]

On Ram drives, They go back loooong before PCIEx; I remember a do-it-yourself ram drive article in an issue of Electronics Now (or was it called Popular Electronics at the time?) back in the days of 286/386 PCs.

[alexander]

you can actually buy them now... Gigabyte manufactures one... unfortunately with WinXP you can only have up to 2 gigs of RamDisk space... Naturaly the border in linux is the total amount of ram you have, so if you have 128 Gigs, you can have, well theoretically, leaving 4 gigs for the system, 124 gig ram drive ;)

 

but you can get a ram disk drive here for example:

http://www.ncix.com/products/index.php?sku=16611&vpn=GC-RAMDISK&manufacture=Gigabyte

 

Thank you for clarifying your terminology cwes, i thought you were talking about some new concept i have yet to hear about, but you were infact not...

Yes i know what caching is, yes i know how it works, both in hard drives and in the microprocessors. It does increase the performance (in hard drives and decrease performance but increase stability in procs), and it does add overhead, negligible compared to the performance boost in HDs. But caching still leads to the same peoblems. Puling out a drive at idle will not damage the FS, whether caching is on or not. The situation i have been describing happens when the drive is pulled out in the middle of being written to. It would be equaly stupid to pull out the drive while its being accessed, whether you have caching or not (hence why physical caching needs to be created on those things). if you are reading from the disk, you can safely pull it out in the middle of the operation, reading does not modify data, however if there is a writing action going on, i would say that perhaps having write caching under some circomstances may be better then not having it.

 

If you are writing to the disk with caching, the cache gets filled up before the contents are dumped on the drive (perfornace on a harddrive is increaded because every time you write you need to move the head to the location of stuff being written to, and that may take up to 10milliseconds, so less you do that, less time you spend on moving the head, hence performance increase), while writing to the disk with no cache will weild the action to be performed as you are telling the thing to write to the disk (in case of hard drive every time to tell it to write, the head moves, 10ms are lost, write finishes, read continues, move again, now new write right after read and so forth and more and more time is lost). If you pull out the drive in a cached system while the cache is filling up FS dont get effected, but writing sessions are longer, so you are more likely to hit one, you stand a somewhat better chance of not having FS corruption then if you are writing to the disk with no caching where write cycles are shorter... But in either case if you are in mid action, you are srewed no matter what... (you also have to realize how bad FAT filesystem is, but here it plays very little role, any fs may get corrupted by such actions (there is no journaling in FAT, and Journaling may help you recover that section of FS on a damaged drive))

Anyways its not a soft mount, its merely the absence of write caching, and i would still recommend unmounting the drive before removing it.

[cwes99_03]

I guess I would still call it a somewhat softmount. Basically if it isn't writing or reading, the disk isn't in use. In that case it is just like an old floppy, but without the inherent problems a floppy had (I've washed and dried 2 different flash drives and didn't have any problem with either, try doing that with a floppy. I didn't even lose any data off of my flash drives.)

 

I don't believe MS said you could remove the drive while reading or writing without problems. They are simply saying that write caching is disabled by default so as soon as you tell it to save something it actually saves it to the device, instead of saving it to the local HD and writing to it only after you have told it you want to remove the flash drive or shut down the computer. When write caching is enabled, pulling the flash drive out without stopping means that a process that was running on the flash drive and on the hard drive didn't finish executing, and you will definitely lose data on your flash drive (everything you haven't allowed to be saved.)

 

Ok, on to the other thing. The I-Ram SATA drive is cool. That's something I hadn't seen before. I know about RAM disks and their usefullness, but as was pointed out, when you shut down the comptuer it gets wiped out. The I-RAM saves the data on the physical ram, but you still lose a bit of performance (not saying it isn't better than a spinning HD) because of the SATA connection, vs a direct copper to processor link on the MBoard that I was imagining with an onboard flash memory device.

 

http://www.tgdaily.com/2006/03/21/32gb_ssd_samsung/

Samsung is planning on replacing traditional HD's with flash-based HD's in mobile computers.

 

As for the 2GB or 4GB limit, perhaps you should read the following. Basically, every computer regardless of the OS is limited by the hardware. If you have a 32-bit processor you are limited to 4GB or physical memory address space. With swapping or page file useage, and some boot switches on your kernel you can adjust how much memory your kernel and any one process can use. With 64 bit architecture the 4GB limit is replaced by a new much higher limit.

 

http://www.brianmadden.com/content/content.asp?ID=143

http://www.msexchange.org/tutorials/Why-64-Bit-Good-E12.html

 

http://support.microsoft.com/kb/268230

http://www.brianmadden.com/content/content.asp?ID=69

http://www.oreillynet.com/windows/blog/2006/03/overcoming_the_windows_2gb_cac.html

 

And most importantly these on Linux RAM availability

http://www.spack.org/wiki/LinuxRamLimits

http://www.theserverside.com/discussions/thread.tss?thread_id=26347

 

Regardless of this however, the flash memory on the MBoard would not be for use as memory in the sense you are imagining (it isn't process accessed memory, though the page file could be). The flash memory would be treated as a hard drive in my design (can I call it a design, I mean I'm no hardware developer) idea, and should be resizeable just like any hard drive (which I recognize still has limits set on it depending on the MBoard, but we are talking TB in size and I have already stated that the sys volume shouldn't need more than 4GB or so to run WindowsXPSP2, with room to spare.) Does anyone reading this know if this idea is feasable and or already in production, other than that IRAM device Alex already suggested.

[alexander]

That link has no relevance to my post, you simply can not run 2.4 with that much ram, rather you can, but the ram will not have a way of being utilized, so you will be stuck with the processor cap of 4 gigs for a 32 bit system (or 2^32 or 4294967296 bytes of ram), 2.6 kernel has this nifty feature(module) called high memory support, it does memory management differently then any sane kernel memory management module (and you can read all about it at http://kerneltrap.org/node/2450 ), that allows you to have and use 64 gigs of ram, never exeeding 4 gig ram limitation of the processor (and if you have $22,000 you can try that out for yourself, just buy yourself a dual opteron server with 64 gigs of ram and play around).

 

As to flash, yes i know that iram device is not flash, but it shocases the speed of nonspinning disks on high speed buses. Actually built in flash would be a lot better, no the iram device does not loose the data, that is what the backup battery is for.

And that Iram device runs on a pci bus, flash can have its own 6gigabit bus much like the memory bus on the PS3...

 

I don't believe MS said you could remove the drive while reading or writing without problems. They are simply saying that write caching is disabled by default so as soon as you tell it to save something it actually saves it to the device, instead of saving it to the local HD and writing to it only after you have told it you want to remove the flash drive or shut down the computer. When write caching is enabled, pulling the flash drive out without stopping means that a process that was running on the flash drive and on the hard drive didn't finish executing, and you will definitely lose data on your flash drive (everything you haven't allowed to be saved.)

 

That is exactly what every presentation of vista i have seen has said "you can pull it out at any moment with guaranteed no data loss" and that only applies when you use the drive as "ram" aka swap

 

And that is not how cache works! it does not spool all you actions and them execute them at unmount point, caching cashes a bunch of actions and then executes the instructions at once, it will wait for all read instructions to finish and then write (in blocks), no caching takes the data and performs the write instruction whether or not the read instruction is being performed, it waits till the end of a tiny read statement, writes and lets read continue after it is done with that particualr write, then next write comes along and similar action is performed, only it takes time to switch from reading to writing, you have to locate your write to position, then when you are done, you have to resume read by locating the last read from position, that is what takes time.

Here is an experiment for you, take a flash drive and plug it in, light goes on it mount (use caching). now take a file and copy it to the disk, wait until the light on the drive goes off and pull the drive out (without unmounting). Now plug it back in, is the file there?

[cwes99_03]

1) yes I know this is how it works with the 4gb limit. That's what the links said. They also pointed out that linux has the same problem on the same intel 32 bit technology. Unless I am missing something here, there's no difference between how much ram can be utilized. That was also covered in the links that I sent you.

 

2) as to the IRAM, the battery is necessary to keep the ram active and hold the data. What happens when the battery dies, you loose the data. Flash doesn't require a battery, right? I didn't think it did, but hey, I might be wrong.

 

3) The IRAM device you showed does not communicate via a PCI bus. It communicates using SATA. The PCI bus is just for power and sharing of processor for the card components. Data transfer uses a SATA channel.

 

4) Could you point me toward a link that says you can pull it out while it is being written to and not lose any data?

 

5) I haven't actually tried it but I have an NTFS file system on mine, so wouldn't the NTFS protect it from total damage. PS - I'm just repeating some of what I've read. I read a bit about caching and they said that when you eject the drive, it writes anything it has cached and then unmounts the drive. When write caching is not enabled, then it treats the drive like a floppy used to be treated. You could eject the floppy at anytime (as long as you weren't opening a file or writing to a file) and not have any problems. When you did eject during a transfer, the computer blue-screened saying that the drive was no longer accessible due to a failure.

 

I think we both need to do a bit more reading about MS and write-caching. Write caching does not refer to using the flash drive as a page file to the best of my knowledge, but I shall do more digging. Also there obviously is no spinning disk and reallignment of the reading head for flash drives, so i don't know how that affects caching on a flash drive.

[alexander]

1) yes I know this is how it works with the 4gb limit. That's what the links said. They also pointed out that linux has the same problem on the same intel 32 bit technology. Unless I am missing something here, there's no difference between how much ram can be utilized. That was also covered in the links that I sent you.

 

do you choose not to follow the links? ok, ok i wont go into a flaming, we are civilized people, but it really seems that what i am talking about is not what you are talking about and we are making 2 slightly different points!

 

even on 32 bit architecture the high memory module enables the support of over 4 gigs of ram and from PPro on, up to 64 GIGs of RAM. ugh, that's what i said last time.... ok, read up on Physical Address Extension here (Follow the link this time please) http://en.wikipedia.org/wiki/Physical_Address_Extension

 

2) as to the IRAM, the battery is necessary to keep the ram active and hold the data. What happens when the battery dies, you loose the data. Flash doesn't require a battery, right? I didn't think it did, but hey, I might be wrong.

 

Have i not said that Flash memory is better?

Actually built in flash would be a lot better

Onboard flash would be amazing

hmm... thought so... NAND flash memory is cheap to manufacture, relies solely on software for bad block management and may come from the manufacturer with bad block already, needless to say that it does not quite operate like a hard drive so it would not go bad quite as easily...

Yes it is already produced and used in such fashion, on various network appliances and such, no i have not yet heard anything about that coming stock on motherboards.

 

3) The IRAM device you showed does not communicate via a PCI bus. It communicates using SATA. The PCI bus is just for power and sharing of processor for the card components. Data transfer uses a SATA channel.

i just found something quick... it would be cool though if it had its own scsi controller on it and did not rely on SATA...

 

5) I haven't actually tried it but I have an NTFS file system on mine, so wouldn't the NTFS protect it from total damage. PS - I'm just repeating some of what I've read. I read a bit about caching and they said that when you eject the drive, it writes anything it has cached and then unmounts the drive. When write caching is not enabled, then it treats the drive like a floppy used to be treated. You could eject the floppy at anytime (as long as you weren't opening a file or writing to a file) and not have any problems. When you did eject during a transfer, the computer blue-screened saying that the drive was no longer accessible due to a failure.

 

NTFS is a horrible file system, yes to some extent it would make it recoverable.

 

I cant beleive that i gotta have FAT32 on my flash drives, because stupid M$ does not support any good filesystems!..

 

Yes with caching when you unmount the volume the final things that were cached will be written, but i have done that experiment about a gazillion times with caching, and i am yet to loose any data...

[cwes99_03]

OK,

 

1) we agree that a NAND based flash drive built into the motherboard would be super fast and an awesome way to load the OS.

CMOS and BIOS pretty much do some of this, only with way fewer processes running at start up right?

 

2) the 4GB limit is still under discussion, as we both are trying to get a grip on what each other is talking about.

 

the 4GB limit of 32-bit architecture (processor v2.4) is resolved with PAE in processor v2.6. So Linux running on a v2.4 intel processor couldn't access more than 4GB either right?

[alexander]

there is one challenge that i see in the flash though, MS would have to redesign their OS so you can have main system folders on one physical drive, while the Applications installed are on another physical drive, while making it seem effortless especialy at installation time... that and build in support for some real filesystems for christ sake! (ext3 and reiserfs3 and 4 at least...)

 

(linux (or any Unix-based os) can use up to 6 (on my systems at least, i know others have more), making it usable in this matter of flash memory (you can put the swap and / on the flash, while /usr, /var and /home may remain on the HD, oh and link /opt to /usr/opt)

 

v2.4 and 2.6 are that i refer to are linux kernel tree versions (they are very different), the Wiki says that PAE is a feature of Intel chips from PPro to current day, but most OSes dont support it (windows, Solaris, BE, OS X and others), and Linux HighMem module does exactly that though. And i think AMD has had some crazy feature like that forever, however it is not clear from the articles i've read so far, thought to add, most of the PCs that I've seen advertised with up to 128 gigs of ram have all had AMDs in them (thought that was weird that while they offer intels with that much ram, they advertise only AMD :confused: ).

[cwes99_03]

Ok, but here I think you demonstrate a misunderstanding. Either that or I seriously don't understand it either.

 

AMD and Intel are chip makers. Thus I accept they are different. What you are telling me is that AMD doesn't have the same 4 gig limit as non PAE intel chips had. Then you and I also agree that PAE Intel chips don't have the same 4 gig limit that non-PAE chips had.

 

Whether PAE has been available since PentiumPro chips or not, I don't know. I'd have to do some searching.

 

v2.4 and v2.6 have nothing to do with the OS, per my understanding. Perhaps you only thought they were strictly Linux. They are versions of Intel chips. This is my understanding due to the websites I posted earlier for you to read, so perhaps you could read them and clarify what they say if I misunderstood them.

 

The windows OS does allow for memory configurations above the 4gig limit, as you yourself have pointed out. So it isn't an incapability of the OS, it is how the OS is created to be limited by microsoft. Since windows 98, ME, 2000, and to some extend XP weren't created to run applications that needed that much memory, why should they allow the use of it? I don't know, this is where my current knowledge becomes fuzzy, because we start talking about architecture, per-process limits, limitation due to hardware, etc. that requires some sorting out.

I'd appreciate a lesson in some of these last things, but before attempting to do so, let's make sure we are both on the same page about the above stuff. Crossed up wires in terminology are counter productive to learning.

[cwes99_03]

Windows OS already can re-locate the program files directory. In fact you can do so when installing many applications. At installation a registry setting is written pointing to where the files are located.

I assumed that a change to the OS would be necessary.

 

BTW, I just had a brain bomb go off. I just realized that installing an OS on flash memory is what happens on all kinds of devices like palm pilots and what not. Why not extend this into the realm of desktop devices?

[CraigD]

Windows OS already can re-locate the program files directory. In fact you can do so when installing many applications. At installation a registry setting is written pointing to where the files are located.

I assumed that a change to the OS would be necessary.

 

BTW, I just had a brain bomb go off. I just realized that installing an OS on flash memory is what happens on all kinds of devices like palm pilots and what not. Why not extend this into the realm of desktop devices?

As I understand it, most PDAs keep their OS and built-in apps (and in newer machines, such as PalmOS 5.4+ ones, the whole memory) in non-volatile (almost (?) always Flash) memory, but copy it into volatile memory when powered up, or just before accessing. Nearly all NVM reads much slower (about 1000 times) than VM, so unless it’s OK for a device to be relatively slow, you’ve got to have some sort of scheme like this.

 

Memory architecture is one of several areas where my expectation of technology diverged dramatically from the reality: In the late 1980s, having seen IBM ES9000 mainframe SRAMs with read and write speeds limited only by how well they could dissipate heat (typically delivering about the equivalent to about a modern 1 GHz FSB), I predicted that, by the mid 1990s, DRAMs would be on the dustbin of history. Although a few all SRAM PCs were made in the early-mid 1990s (all with 1:1 memory bus:CPU clock speed), they remained pricey, high-performance oddities. Ah, but they were beautiful – with no caches or pipelines, they were almost completely deterministic (until slow, uncertain things like disks).

 

Now, with the prospect of Flash on top of DRAM on top of CPU cache, and EMC/Shark Drive farms for storage, memory architecture continues to evolve in almost the exact opposite direction from what I predicted!

[alexander]

Whether PAE has been available since PentiumPro chips or not, I don't know. I'd have to do some searching.

if you follow my wiki link, or at least go to wikipedia and search for PAE, you will find all that info available to you :shrug:

actualy yes thinking about it, there is a 4 gig/process limitation on PAE compattible procs... in 32bit x86 arch

Perhaps you only thought they were strictly Linux. They are versions of Intel chips. This is my understanding due to the websites I posted earlier for you to read, so perhaps you could read them and clarify what they say if I misunderstood them.

Linux 2.4.x does support greater then 4GB of physical memory

 

Actually while reading the High memory module about, it seems that PAE actually used to support up to 4 gigs of ram, back when it cam out on PPros, now however with P4s it alows you to have up to 64 gigs of ram on 32 bit architecture. You can read more on how the memory is managed here:

http://kerneltrap.org/node/2450

 

Trust me its pretty darn crazy...

 

BTW, I just had a brain bomb go off. I just realized that installing an OS on flash memory is what happens on all kinds of devices like palm pilots and what not. Why not extend this into the realm of desktop devices?

lol, i said that when i told you about flash-running appliances :hihi: There are firewalls, routers, web-servers, nas devices and so forth that run flash, outside of palm tops and things of consumer nature

 

And i think it is not mainstream because it would require M$ to change their installer, you see, M$ uses one partition to install their OS, while you can install programs on different partition, you can not make the Program Files folder be permanently positioned on another drive at install time and have the install be seamless, you can not split the Windows folder across multiple drives, windows is not really a muli partition-friendly OS...

 

Oh and another quick thing, Windows CE is trivialy different from WinXP (i deal with CE a lot at work, they are really different)

 

I think that if M$ changes their installer to use multiple partitions (and multiple partitions can make the OS safer by separating trivial areas and limiting access to them to certain users/groups) then building in flash with extremely fast bus speeds will actually be useful and used to increase system performance...

[cwes99_03]

Thanks, for the pat on the back for an idea that may not be originally mine. I agree that if MS would update their thinking to include a flash based OS volume on the MB and would work with MB manufacturers to get it happening that would be awesome.

 

I agree that you can't just move the Program Files folder. My point was that if you could install a program anywhere on any drive you want and windows could still access it then the program files folder was really kind of useless other than for those apps that require you to store it there (because they don't like allowing custom installation.) Would you agree that 4GB for this flash volume would accomodate any OS (with this modification to create program directories on another volume.)

I know they make 32GB flash drives now, but would 4gb be enough (as it certainly is cheaper right now) since we are only worrying about a stable OS and not your files.

[alexander]

I think it should be enough, any OS should fit in 4 gigs, i mean if we are talking packages, and windows does not distinguish between an OS and a web browser and notepad and a window manager (although that can stay),

 

you should accomodate 4 gigs pretty well, its enough to have windows installed (full install) i have however seen widndows files to get over the size, but if it was used for OS purpose only, no logs or junk files, then i think that it would accomodate any OS to come too, even if you include the window manager with it, just that you cant do the same stupid stuff that M$ does and install a gazillion applications that noone will ever use together with an OS on the same drive without giving anyone ability to delete those stupid programs...

[cwes99_03]

Since this all began as a vista rant, how big is the current vista OS install? I heard they are sending it out on DVD's because it doesn't fit on a CD.

[alexander]

a fresh vista install uses up about 8.5Gigs from what i have read so far... :shrug:

[cwes99_03]

Yah, that is incredulous. But then I think that is for the ultimate media center version which has tons of add ons.

I currently have a computer running ultimate media rc2 and I have to say it is quite a large install.