On May 11, 2010, at 5:23 PM, John Gilmore wrote:

> There's a classic Unix problem with distribution of disk access times
> that relates to how older Unixes did "sync" -- every 30 seconds there was
> an instant traffic jam at the interface to the drives.  This has been
> studied to death; here are some assorted papers:
> 
>  http://www.cis.upenn.edu/~jms/usenix.pdf
>       Has graphs from old DEC hardware that look similar to XO-1.5.
> 
>  http://www.usenix.org/events/osdi06/tech/nightingale/nightingale.pdf
>       Good references section.
> 
>  http://www.eecs.umich.edu/Rio/papers/chen96_2.pdf
>       Studies ext3 performance in depth.
> 
> Most such studies looked at read() times because the usual workload has
> far more reads than writes.  Odd delays in write() times are interesting.

Writes in UNIX are typically asynchronous --- get the data into a memory
buffer and the program can continue along its merry way.
I strongly suspect that writes through the SD system are forced to wait
for the media to complete the write.

wad

_______________________________________________
Devel mailing list
[email protected]
http://lists.laptop.org/listinfo/devel

Reply via email to