[luau] No hard drive, only compact flash card
MonMotha
monmotha at indy.rr.com
Sun Jul 6 22:09:01 PDT 2003
Hawaii Linux Institute wrote:
...
> Second, the access time of flash memory is measured in
> nanoseconds, thus, there is no latency compared to hard disc, which
> typically has an access time measured in miniseconds.
Milliseconds you mean. But I keep forgetting about this BIG one. In situations
where sustained throughput is low, but random access are high, CF cards may
actually be FASTER. A flash read cycle always takes the same amount of time
(generally on the order of 30-120ns for NOR flash, not sure about NAND whcih
many of the larger CF cards may use), no matter where you're accessing relative
to where you just accessed. Hard drives are great for sequential access, but
really suck on highly random small reads (say only a few bytes) because they
have a moving head that has to seek. Seek times for consumer IDE hard drives
seem to be running around 5-20ms these days.
>
> Some new generations of USB 2.0 compliant compact flash sticks can write
> faster than read (and can read and write simultaneously). While the
> write speed is currently maxed out below 10 MB/s, the technology is
> evolving very rapidly. If a market exists, a single channel USB 2.0
> compliant CF disc may reach 60 MB/s. This is similar to the speed of
> ATA/66. Eventually, who knows, there may be multiple-channel CF discs.
> And that will be a totally different story.
I'd like to know how this is technologically possible. Assuming NOR flash (I
will admit up front that NAND flash may be MUCH faster) with a total cycle time
(including address set up and data read strobe) of 30ns (an estimate, but the
range is big, I think this is on the low end) with a 32bit wide data bus (mind
you, the CF data bus is only 12 bits wide), and assuming that the bus and CF
cycles are always in sync (they're not), we have a max theoretical of abotu
128MB/sec. This is assuming a pretty fast chip with no overhead and NO ERASE
CYCLES (which take a LONG time). Writes generally involve erase cycles (unless
they've gotten smart and erase things during idle time). It is posisble, but
VERY unlikely that you would hit 60MB/sec. Are you sure this isn't the
theoretical bandwidth of the bus? I'm not denying that it exists, just
questioning the feasbility of it given current flash technology). I mostly work
with just the ATA ones, not the USB ones as well (because they're cheap, and I'm
using them as ATA devices :).
>
> Wayne
--MonMotha
More information about the LUAU
mailing list