What is Digg?Sponsored by Best Buy
PS3 view!
bestbuygiftadvice.com - The Best Buy(R) Twelpforce(TM) Carolers sing about the impressive Blu-ray(TM) capabilities of the PS3.
88 Comments
- Error601, on 10/10/2007, -9/+65The word hybrid should be banned from all marketing literature.
- NSMike, on 10/10/2007, -1/+30I'm not sure that would be a problem... With the right control scheme in place, all writing would be done directly to the disk, while everything that is frequently loaded is pre-cached in this RAM. Only thing that would happen is the pre-cached data would be lost, and if the system is smart, nothing important would be there in the first place.
- rnreekez, on 10/10/2007, -1/+29do you prefer mashup? haha....
- alanrice38181, on 10/10/2007, -8/+29makes you what one?
want would that mean exactly? - FortyCaliber, on 10/10/2007, -3/+20Want are you talking about? I would what one too!
- fivestarsoul, on 10/10/2007, -1/+16use "significantly" instead of highly, and you're golden.
- theratdotus, on 10/10/2007, -1/+14the way i know it is the os would be stored on this extra cache, so when you first boot, it writes fast to ram cause its already called. if it wasnt on this extra ram chip, the harddrive would have to spin to get all the needed data to the ram. we couls skip all this with solid state HDs. San disk makes a 32 gb sd card, so we are getting close.
Damn, 32 gb sd card and i read about a 1 mhz processor that only takes one watt to run. combined with a flexible lcd we have a mega small, ultra usable computer that fold up into my *****. - wiifm69, on 10/10/2007, -6/+19This would be great ... until there is a power cut
- pintomp3, on 10/10/2007, -0/+13how about crossbreed?
- gweedo767, on 10/10/2007, -0/+11Somone has never used RAID controllers with large memory cache's before I take it? They have batteries my friend to store the data in case of this occuring.
- Hello1024, on 10/10/2007, -0/+10Can anyone explain why exactly we're putting the cache in the HD? We should put the same amount of cache as extra system memory. System memory can be accessed much faster than even HD cache, and decisions as to what to put in the cache can be made much mre intelligently as the OS has a much better understanding of file system structures, running programs, file load order etc. than the HD could ever have.
Sharing main system ram also means the cache can be dynamicly allocated by the OS kernel - so when memory isn't being used for anything else the cache could be up to 80% of system RAM to speed up the next application launch a lot more.
There's no significant overhead by doing this because te disk uses DMA anyway, and the relative data rate is low compared to the memory bandwidth so it wouldn't slow down the CPU.
The only significant disadvantage is in the case of a power faliure or system crash, data in the write cache might be lost - but in most hard drive caches it would be lost as well. - edmcguirk, on 10/10/2007, -0/+6I don't understand. Is disk cache significantly better then system RAM?
Admittedly, disk cache probably uses cheaper RAM and up until recently it was difficult to get more than 2gig of usable system RAM.
But wouldn't it just be better to have more system RAM? Are there any disks that can deliver data faster than bus speed? With a fast bus and multiple CPUs it just seems that money would be better spent on system RAM. - inactive, on 10/10/2007, -0/+6I would imagine since it is a 2.5" drive in a 3.5" enclosure that the noise would be significantly reduced. Just a guess on my part, though.
- Ramble, on 10/10/2007, -0/+6Why would there be a sound increase? It's just RAM.
- inactive, on 10/10/2007, -0/+6The point is, you're supposed to put your OS on the drive. So you only need around 40 GB anyway. Apps/software/games go on another drive.
- yodaj007, on 10/10/2007, -0/+5Male rams are especially noisy during mating season.
- inactive, on 10/10/2007, -0/+5"Hybrid" is better than "mashup" and "crossbreed", that's for sure.
- underthelinux, on 10/10/2007, -1/+6any word on the noise/volume level? I want a shhhh pc.
- taintedzodiac, on 10/10/2007, -0/+4Should be possible to nullify any sound increase. All the part in these are already elsewhere in your computer.
- OrangeTide, on 10/10/2007, -0/+4For those wondering what the point of it is, the value is the battery backed part. The extra cache would mainly benefit a bad OS, a good OS can be smarter than the harddrive is when it comes to caching. Having it battery backed makes transactional operations faster because you can commit the data without having to wait for the heads to actually write it to the media, you can let it sit in the battery backed ram and count on the data being committed even if you unplug the drive or the OS crashes or whatever.
In the storage appliance industry we tend to buy a PCI card that has battery backed RAM, then use it as a journal for transactions. Which we do a second level commit to the disk drives according to a schedule, to free our cache/journal memory. This seems to just do the same thing in a small modular form factor with some automation added in for good measure. - dAbReAkA, on 10/10/2007, -0/+4because it would probably read more often (like, when it's idle, trying to fill that RAM-cache)
it's the same with superfetch.. increased hard drive activity till it fills up my 1.5 gig RAM.. and with that noisy hard drive of mine, it's hell.. - inactive, on 10/10/2007, -1/+5Are you someone who would reply the same thing to the statement "Soon to be released: hooblijoobly laser xenon encrustred ddxr4 harddrives?"
- koko775, on 10/10/2007, -0/+3The CPU cache and memory system, for one thing, is limited on 32-bit systems to 4GB. In the common case, you would get no performance increase BECAUSE of the complexity exposed to the OS. This would mean that the extra memory given to the OS, barring some sort of virtual memory system architected to accomodate this memory, would either be left unused or require the OS to be designed around a special virtual memory system. On the other hand, since you can assume that stuff on a hard drive (i.e. opened media files or documents or whatever) are going to be accessed more than just once while they're open, i.e. opening and saving a file, viewing media files, listening to music, you can cache that on initial read onto the memory and save the hard drive from having to spin up when the file, or, more accurately, nearby disk sectors, are re-accessed. This is also the reason it's good to defragment, unless I'm mistaken -- contiguous files means better HD cache usage, as well as physical locality on the disk platter.
Besides, who's to say that the HD wouldn't also cache the memory swap file? - atbnet, on 10/10/2007, -0/+3I paid $150 for a 74 GB WD 10k drive. Put the OS and the games you want to load fast on there and you're golden. Movies, TV shows and music go on the 750 GB drive.
- incognit0, on 10/10/2007, -0/+3It's a write-through cache so theres really nothing to worry about...
- itseffinkasey, on 10/10/2007, -2/+5Cool so I'll pay $136 for a super fast (65 gig) 80 gig hard drive! Sounds kinda small...
- HeyArnold, on 10/10/2007, -0/+3or crossover?
- Ramble, on 10/10/2007, -0/+3Your argument only works on pre-Vista systems. Superfetch effectly works the same way this would, only you can expand the space.
- Firehed, on 10/10/2007, -0/+3Considering that the rest of your system (excepting the optical drive) is running somewhere around a hundred times faster, a huge speed improvement would be worth that cost to many people.
My first impression here was "about friggen time". We can't easily make platters spin at 50,000 RPM, so why not at least store as much randomly-accessed data as possible in something with almost zero seek time and bandwidth limited by the interface? - jdoo, on 10/10/2007, -0/+2Your operating system already does this. In fact, the OS is much smarter about the data it needs to keep cached since it has more knowledge about what you are doing on your computer than a disk could possibly be aware of. Just go buy another 1GB of ram and a larger hard drive.
- Steeple, on 10/10/2007, -0/+2photoshop scratch disks can NEVER be too fast
- pintomp3, on 10/10/2007, -1/+3more importantly, seeking to the money shot will be even faster.
- mikm, on 10/10/2007, -0/+2What does email have to do with anything?
- digitalarcanum, on 10/10/2007, -0/+2I'm agreeing with Duder. You're all acting like *****, like moving from ATA to SCSI would be a bad thing. better loadtimes better throughput, and support for more devices in a chain.
- zoom1928, on 10/10/2007, -2/+4You are correct if you use a good OS. If you use one that consistently makes bad decisions on caching like Windows, then you're better off having a safe fall-back on the drives where you know that the extra cache is needed. This discussion mirrors the one about 15 years ago when caching RAID controllers first become available for reason prices. The RAM on the controller wasn't worth the extra in the general case, but when Windows NT came-out, it suddenly became worth every penny. I don't know if Microsoft has ever gotten around to fixing NT's major problems with sizing the disk cache versus system memory, but knowing the other problems that Microsoft hasn't fixed in over a decade, I would guess there's still major problems.
- OrangeTide, on 10/10/2007, -0/+2aren't those 7200rpm perpendicular recording hard drives getting better throughput than the 10k rpm raptors (which use older technology)?
- Ramble, on 10/10/2007, -0/+2Only if it's cached. For the vast majority of data on the disk it will be the same. It is only as fast as the slowest part.
- mastertop, on 10/10/2007, -0/+2a 1mhz processor that takes one watt ? Wow, that sucks
- HairyFotr, on 10/10/2007, -0/+2The disk probaly doesn't do this, but: it could have a feature that would analyze which data gets loaded at bootup and start filling that data into the cache the moment you turn on the computer. Even if the bootup sequence changes completely there is no harm in having useless data in the cache and the software could be made to adapt to changes.
- keyboardduder, on 10/10/2007, -0/+2and the devices dont have to use an IRQ Every time they need something, the chain controllers have their own processors and they use less CPU time too. Its a wonder why they dont use SCSI in consumer applications much. (post 1990 apple os lol)
- Hello1024, on 10/10/2007, -0/+2Remember the company (PCChips I think) that embedded code into it's motherboards BIOS's to modify windows OS code on the fly during windows installation to make it download and install a program of their choice on first boot after installation of windows.
I suffered their bios, and spent ages reinstalling windows using known good disks and clean formatted hd's to find that they were still infected with adware - only to discover the adware came in the PC's hardware!
Luckily that bios only affects windows 98SE - install any other OS and you're safe.
Even though, it's a pretty impresive feat of programming to be able to infiltrate an OS from such a low level, and an attack vector nobodys ever thought about. - jcaino, on 10/10/2007, -3/+4nope, you're being dugg down for being a pretentious *****.
- manitoba98xp, on 10/10/2007, -0/+1The point is, this could be done in the hard disk, too (closer to the disk, so more transparent, plus more caching memory in general is good.)
- Hello1024, on 10/10/2007, -0/+1firmware on pretty much all consumer hardware is closed source - this is no different really. For the company it's probably cheaper and easier to develop on a microkernal rather than write all the underlying stuff from scratch.
- SteveMax, on 10/10/2007, -0/+1"a SATA-II drive"
Make/model/capacity? This could be the problem. - Hello1024, on 10/10/2007, -0/+1flash drives have good random write performance with the right filesystem, provided the block being written to is pre-erased.
- Hello1024, on 10/10/2007, -1/+2Even if the HD has knowledge of the FS structure, it still can't match kernel managed caching - for example the kernel knows what files will be needed in advance of being required (for example during boot or application startup where the read requirements are fairly predictable). Also, the OS knows prioritys of various tasks for IO, and can read data into the cache with that in mind, (so creating indexing data is a low priority task, so othe applications data gets to use more cache space)
- neutrascrub, on 10/10/2007, -1/+2i prefer lesbian pron, thank you very much
- inactive, on 10/10/2007, -0/+1Unfortunately you're only right if the software loads the next segment into the memory before you complete the current one. Then you're looking at a nice and easy memory to memory transfer.
- Hello1024, on 10/10/2007, -0/+1Complexity doesn't mean poor performance - look at things like memory paging/memory management/dynamic memory allocation as an example of that - all complex procedures, but very well tuned so they're super fast. (although I agree with a bit of cpu hardware support).
On your next point, 32 bit is a problem - effectively by using a massive HD cache, you are reducing this problem a bit by offloading data that could be in RAM into the cache - and in the case of the swapfile it's actually a pretty direct addition to RAM, except there is significant overhead in copying from system ram to disk cache - there are many orders of magnitude between RAM access times and disk interface access times, added to the fact that data has to be accessed from the disk in blocks. -
Show 51 - 88 of 88 discussions
© Digg Inc. 2009
— Content created and posted by Digg users is