Re: [math-fun] one crazy idea
I don't know about seek times, but one of the major problems with almost all storage systems is that the time to _process every byte_ continues to go up, because storage density is going up faster than transfer rates. Multiply the fastest transfer rate of your disk drive by its capacity. You'll be astounded at how long it can take to read the whole thing. If you have a large (64GByte) flash drive, calculate how long it will take to _write_ the whole thing. These times are lengthened by another order of magnitude if the data consists of small files (e.g., Unix/Linux). (Actual transfer times are usually considerably slower than advertised, due to the inability of OS's like Windows to do proper write caching for flash drives.) At 05:30 PM 7/17/2013, Bill Gosper wrote:
This reminds me of my dumb idea to reduce disk seek times with multiple heads per surface. Surface is cheap.
[this is OT, but I can't keep my mouth shut...] * Henry Baker <hbaker1@pipeline.com> [Jul 18. 2013 17:48]:
[...]
If you have a large (64GByte) flash drive, calculate how long it will take to _write_ the whole thing.
Even if you have to stick to measly USB (ver. 3), you should get > 120 MByte/sec (I think 400 MB/sec is already there, not even very expensive). SSD disks were at somewhere like 500 MB/seclast time I checked (a year or so ago).
These times are lengthened by another order of magnitude if the data consists of small files (e.g., Unix/Linux).
Unix meaning "many small files" ceases to be true by today's standards: my (for Unix dinosaurs ridiculously populated) /etc/ is totaling at 30 MBytes (in 3530 files). Modern file systems have clever ways of handling the "many small files" scenario, even the restriction "one block belongs to one file" isn't always true for certain file systems.
(Actual transfer times are usually considerably slower than advertised, due to the inability of OS's like Windows to do proper write caching for flash drives.)
I'd be surprised if that's still true with Win7 or Win8 (but not overly surprised...).
At 05:30 PM 7/17/2013, Bill Gosper wrote:
This reminds me of my dumb idea to reduce disk seek times with multiple heads per surface. Surface is cheap.
That has been in the past, but just 2 heads (IIRC) with magnetic disks, and several heads for a certain "72-speed" CDROM drive that could do this without spinning the disk into explosive destruction.
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com http://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
participants (2)
-
Henry Baker -
Joerg Arndt