Impact of solid-state drives
in [Unix Programming]
|
Prev: gethostbyname doesn't accept IPv6 addresses
Next: 7 x = ."It's" IF ." " ELSE ."not " THEN ."equal"
From: kenney on 7 Mar 2010 06:12 In article <hmruc0$g04$1(a)USTR-NEWS.TR.UNISYS.COM>, timcaffrey(a)aol.com (Tim McCaffrey) wrote: > And since it is used for paging, you don't need to worry about data > going away when the power goes off. Be great for temp files (browser > cache, etc) as well. Congratulations you have just reinvented the RAM drive. The concept of using RAM as a virtual drive goes back to the 8 bit period as it was a much easier way to use extra RAM than page switching and drives were a lot slower then. There was also a lot of use of overlays t fit programs and data into 64k. Something like the Amstrad PCW had 256k of RAM with most of it being used as a RAM disc. I can remember with a data base program on a Video Genie how slow things got when the data set got too big to fit into memory. Ken Young
From: Tim McCaffrey on 7 Mar 2010 11:06 In article <7uadnf78saeJFw7WnZ2dnUVZ8k2dnZ2d(a)giganews.com>, kenney(a)cix.compulink.co.uk says... > >In article <hmruc0$g04$1(a)USTR-NEWS.TR.UNISYS.COM>, timcaffrey(a)aol.com >(Tim McCaffrey) wrote: > >> And since it is used for paging, you don't need to worry about data >> going away when the power goes off. Be great for temp files (browser >> cache, etc) as well. > > Congratulations you have just reinvented the RAM drive. The concept of >using RAM as a virtual drive goes back to the 8 bit period as it was a >much easier way to use extra RAM than page switching and drives were a >lot slower then. There was also a lot of use of overlays t fit programs >and data into 64k. Something like the Amstrad PCW had 256k of RAM with >most of it being used as a RAM disc. > > I can remember with a data base program on a Video Genie how slow >things got when the data set got too big to fit into memory. > I'm well aware of RAM drives that use main memory, I'm talking about building something that is far faster than a Flash disk, but (hopefully) at a lower price point than main memory RAM. - Tim
From: Stefan Monnier on 7 Mar 2010 15:07 > I'm well aware of RAM drives that use main memory, I'm talking about > building something that is far faster than a Flash disk, but > (hopefully) at a lower price point than main memory RAM. Is that at all possible? IIUC current DRAM is mainly driven by "dollars per megabyte" already (i.e. speed is a very secondary concern), so I wonder if there's really much we can do to reduce its price. Stefan
From: Thomas Womack on 7 Mar 2010 17:44 In article <jwvmxyjlwyp.fsf-monnier+comp.arch(a)gnu.org>, Stefan Monnier <monnier(a)iro.umontreal.ca> wrote: >> I'm well aware of RAM drives that use main memory, I'm talking about >> building something that is far faster than a Flash disk, but >> (hopefully) at a lower price point than main memory RAM. > >Is that at all possible? IIUC current DRAM is mainly driven by "dollars >per megabyte" already (i.e. speed is a very secondary concern), so >I wonder if there's really much we can do to reduce its price. You can't get much cheaper than current DIMM sticks, but you can probably get bulk memory substantially more cheaply than having it be the main memory of whole computers: attach it to controllers which are not whole quad-core CPUs. Be prepared to have two dozen controllers each with half a dozen DIMMs attached, and outwardly connected to some protocol designed to be the cheapest transcievers available at 1GB/sec; have a master communicate with whatever controller is connected to the memory it wants. Though demand for this probably won't be enough to make the custom controllers and motherboards substantially cheaper than $100 Athlon64 chips on $100 motherboards talking to 8G of DDR2 bought for $200, and a $125 hypertransport HCA switching through a $750 8-port hypertransport switch. Tom
From: Tim McCaffrey on 7 Mar 2010 20:52
In article <VJk*tSq5s(a)news.chiark.greenend.org.uk>, twomack(a)chiark.greenend.org.uk says... > >In article <jwvmxyjlwyp.fsf-monnier+comp.arch(a)gnu.org>, >Stefan Monnier <monnier(a)iro.umontreal.ca> wrote: >>> I'm well aware of RAM drives that use main memory, I'm talking about >>> building something that is far faster than a Flash disk, but >>> (hopefully) at a lower price point than main memory RAM. >> >>Is that at all possible? IIUC current DRAM is mainly driven by "dollars >>per megabyte" already (i.e. speed is a very secondary concern), so >>I wonder if there's really much we can do to reduce its price. > >You can't get much cheaper than current DIMM sticks, but you can >probably get bulk memory substantially more cheaply than having it be >the main memory of whole computers: attach it to controllers which are >not whole quad-core CPUs. Be prepared to have two dozen controllers >each with half a dozen DIMMs attached, and outwardly connected to some >protocol designed to be the cheapest transcievers available at >1GB/sec; have a master communicate with whatever controller is >connected to the memory it wants. > >Though demand for this probably won't be enough to make the custom >controllers and motherboards substantially cheaper than $100 Athlon64 >chips on $100 motherboards talking to 8G of DDR2 bought for $200, and >a $125 hypertransport HCA switching through a $750 8-port >hypertransport switch. > Well, you could go the other way. You could optimize main memory for speed, and not worry so much about the price or density. Then the break even point moves up. I'm not saying this is practical :) -Tim |