[Haskell-cafe] Parallel weirdness

Brandon S. Allbery KF8NH allbery at ece.cmu.edu
Sat Apr 19 12:31:44 EDT 2008


On Apr 19, 2008, at 11:53 , Murray Gross wrote:
> 2. You need to account for I/O buffering (not only by your OP  
> system in RAM, but by your disk controller)--after the first set of  
> I/O operations, your data may be in buffers, so subsequent uses may  
> retrieve data from buffers rather than from the disk itself.  
> Similarly, you also have to take into account paging and cache  
> issues, which could make the first run much slower than immediate  
> subsequent runs.

Note also that, unless you use SCSI or very high-end SATA drives,  
they ignore requests to disable buffering.  (References on request,  
you can probably find them by poking around http://www.pdl.cmu.edu/.   
Short summary:  consumer drives are optimized for benchmarks, not for  
data safety.  This is why early 32-bit Windows releases often lost  
data on shutdown until the shutdown was modified to sleep for 10-15  
seconds.)

-- 
brandon s. allbery [solaris,freebsd,perl,pugs,haskell] allbery at kf8nh.com
system administrator [openafs,heimdal,too many hats] allbery at ece.cmu.edu
electrical and computer engineering, carnegie mellon university    KF8NH




More information about the Haskell-Cafe mailing list