joseph@rutgers.rutgers.edu (Seymour Joseph) (04/18/91)
In article <1991Apr17.143801.22825@m.cs.uiuc.edu> bazyar@ernie (Jawaid Bazyar) writes: > The RAMfast cache is what is termed as "write-through", i.e. any data >written to the card is immediately written through to the disk. Jawaid, I own a RAMfast SCSI card, and your statement conflicts directly with my experience, and my discussions with CV Technologies. When I write data to disk, there is often a delay between when the program starts writing data, and hard disk activity. The cache on the Ramfast is not a write-through cache. The card is Very fast so the delay is never more than a second or two, but it is there. Seymour
meekins@tortoise.cis.ohio-state.edu (timothy lee meekins) (04/18/91)
In article <Apr.17.16.47.15.1991.9916@aramis.rutgers.edu> joseph@rutgers.rutgers.edu (Seymour Joseph) writes: >In article <1991Apr17.143801.22825@m.cs.uiuc.edu> bazyar@ernie (Jawaid Bazyar) >writes: >> The RAMfast cache is what is termed as "write-through", i.e. any data >>written to the card is immediately written through to the disk. > > >Jawaid, > >I own a RAMfast SCSI card, and your statement conflicts directly with my >experience, and my discussions with CV Technologies. When I write data to >disk, there is often a delay between when the program starts writing data, and >hard disk activity. The cache on the Ramfast is not a write-through cache. >The card is Very fast so the delay is never more than a second or two, but it >is there. > >Seymour Yup, you goofed Jawaid. The RamFast has a "write-deferred" cache. When you write a block of data is stored in a cache and execution is immediately returned to the application. While the GS is off doing something else, the RamFast will begin writing the cache to disk in the background. In other words, it defers the writing until later. This allows for very fast write operations. -- +---------------------------S-U-P-P-O-R-T-----------------------------------+ |/ Tim Meekins <<>> Snail Mail: <<>> Apple II \| |> meekins@cis.ohio-state.edu <<>> 8372 Morris Rd. <<>> Forever! <| |\ timm@pro-tcc.cts.com <<>> Hilliard, OH 43026 <<>> /|
asslk@acad2.alaska.edu (Kingry Shane L) (04/18/91)
In article <Apr.17.16.47.15.1991.9916@aramis.rutgers.edu>, joseph@rutgers.rutgers.edu (Seymour Joseph) writes... >In article <1991Apr17.143801.22825@m.cs.uiuc.edu> bazyar@ernie (Jawaid Bazyar) >writes: >> The RAMfast cache is what is termed as "write-through", i.e. any data >>written to the card is immediately written through to the disk. > > >Jawaid, > >I own a RAMfast SCSI card, and your statement conflicts directly with my >experience, and my discussions with CV Technologies. When I write data to >disk, there is often a delay between when the program starts writing data, and >hard disk activity. The cache on the Ramfast is not a write-through cache. >The card is Very fast so the delay is never more than a second or two, but it >is there. > >Seymour I also own a RamFAST/SCSI card alng with a ZIP/GSX1600 (not that it should probably make any difference..). I have noticed that there is often a dealy when writing to the disk, but the delay does NOT tie up the computer. Basically, as in example, you tell the compter to save a file to the disk. The computer (or so, the GS). Now, Appleworks will say that the file is saved to disk, and it will let you continue with whatever else you were doing. Then, about a second later, the info will be written to disk. If you do a ProSEL backup (or a restore for that matter) then a group of data - several files, wil lall be written to the disk at once. Basically, as I understand it, this s one of the purposes of the RamFAST/SCSI system. IT is increadably neat, and very fast. So, I have made a theory: 1) The RamFAST/SCSI will write to the disk if it doesn't recieve any data to read from the computer for awhie. e.g., you send the card a 20K file, and continue with your work. It writes it when it wants to 2) THe RamFAST will write to the computer when its RAM is full (256K). My guess is that it uses both. Either way - it is really worth the money!
taob@pnet91.cts.com (Brian Tao) (04/19/91)
From asslk@acad2.alaska.edu (Kingry Shane L): > 1) The RamFAST/SCSI will write to the disk if it doesn't recieve any > data to read from the computer for awhie. e.g., you send the card > a 20K file, and continue with your work. It writes it when it wants > to > 2) THe RamFAST will write to the computer when its RAM is full (256K). > > My guess is that it uses both. Either way - it is really worth the money! I would much rather have the option of configuring the RAMFast with a write-through cache. The Quantum HD I'm using is very fast, so there won't be much of a delay anyway. My original beef has to do with the speed of the writing, however. If the RAMFast does wait for the 256K to fill up before writing it out, it takes an awfully long time (relatively speaking) for it to fill up. I suspect that perhaps DMA only works in ONE direction: from RAMFast to GS memory. It looks like the GS has to do all the data transfer from RAM back to the SCSI buffer, which is why it can't do high-speed writes to the drive. Too bad. Brian T. Tao *B-) | t569taob@bluffs.scar.utoronto.ca | "Though this be U of Metro Toronto | - or - | madness, yet there Scarberia, ON | taob@pnet91.cts.com | is method in 't."
asslk@acad2.alaska.edu (Kingry Shane L) (04/22/91)
In article <623@generic.UUCP>, taob@pnet91.cts.com (Brian Tao) writes... > > I would much rather have the option of configuring the RAMFast with a >write-through cache. The Quantum HD I'm using is very fast, so there won't be >much of a delay anyway. My original beef has to do with the speed of the >writing, however. If the RAMFast does wait for the 256K to fill up before >writing it out, it takes an awfully long time (relatively speaking) for it to >fill up. I suspect that perhaps DMA only works in ONE direction: from RAMFast >to GS memory. It looks like the GS has to do all the data transfer from RAM >back to the SCSI buffer, which is why it can't do high-speed writes to the >drive. Too bad. > I am also using a Qunatum HD (Q105 PRO)... What I meant was that if the controller card recieves 256K worth of data in say, second or two, or if it realizes that the GS is sending it big files, then it writes 256K or so at a time. This is a hell of a lot faster then writing it as it gets it interms of multiple files and such... As fas as reading goes, I think the maximum transfer rate with a Quantum is 1meg/sec..
taob@pnet91.cts.com (Brian Tao) (04/22/91)
From asslk@acad2.alaska.edu (Kingry Shane L): > What I meant was that if the controller card recieves 256K worth of data > in say, second or two, or if it realizes that the GS is sending it big > files, then it writes 256K or so at a time. But why is it taking so long to fill up 256K? If the GS is writing out a big file, I expect it to DMA a block of memory to the RAMFast buffer. So instead of a 1 or 2-second delay, it would be writing out a FULL buffer every half-second or so... Brian T. Tao *B-) | t569taob@bluffs.scar.utoronto.ca | "Though this be U of Metro Toronto | - or - | madness, yet there Scarberia, ON | taob@pnet91.cts.com | is method in 't."
whitewolf@gnh-starport.cts.com (Tae Song) (04/29/91)
|> What I meant was that if the controller card recieves 256K worth of data |> in say, second or two, or if it realizes that the GS is sending it big |> files, then it writes 256K or so at a time. | | But why is it taking so long to fill up 256K? If the GS is writing out a |big file, I expect it to DMA a block of memory to the RAMFast buffer. So |instead of a 1 or 2-second delay, it would be writing out a FULL buffer every |half-second or so... While the CPU runs at 2.8Mhz, and the DMA controller at 10Mhz, There is still a very large bottleneck. Anything going through the slots to RAM has a 1Mhz bottleneck, no matter what. The controller cuts down as much overhead as possible and that's about it. What's the argument about anyways... The RamFAST (80.9MB Seagate) is faster than my RAM Disk, in throughput. It might be faster when you do a statistics analysis of how fast they are, but the throughput is what counts, and it beats the pants off any disk system, that I can think of for the GS, so do something more productive like writting a Norton Utilities for it or something... geez no body seems to appriciate the marvel of engineering that RAMFast is as if they could do something better... a feat comparable to surviving a fall the Twin Towers... BOTH of them. INET: whitewolf@gnh-starport.cts.com UUCP: crash!gnh-starport!whitewolf ARPA: crash!gnh-starport!whitewolf@nosc.mil