#8313 NORM 9.1.0: Can OFW save-nand be made to go faster?

[Zarro Boogs per Child]

#8313: Can OFW save-nand be made to go faster?
-----------------------------------+----------------------------------------
   Reporter:  thomaswamm           |       Owner:  wmb at firmworks.com                
       Type:  enhancement          |      Status:  new                              
   Priority:  normal               |   Milestone:  9.1.0                            
  Component:  ofw - open firmware  |     Version:  Development build as of this date
 Resolution:                       |    Keywords:                                   
Next_action:  communicate          |    Verified:  0                                
  Blockedby:                       |    Blocking:                                   
-----------------------------------+----------------------------------------

Comment(by wmb at firmworks.com):

 I just made a small change to the save-nand code that speeds thing up a
 little - the code that scans to see if blocks are erased (so it can avoid
 storing them) was slower than it needed to be.  Optimizing that routine
 knocked off a couple of minutes in the case where the NAND is half full.
 It wouldn't make any difference for a full NAND.

 With that change, I stop-watched a save-nand at 6 minutes, 50 seconds for
 a 436 MB output image.  That's on my "gold standard" Sandisk Cruzer Mini
 USB key, writing to a factory-format FAT filesystem.

 The system I did this on is a test unit that has a 2 GB NAND chip, so it
 had to read twice as many NAND blocks as it would on a normal XO (the
 timing parameters of this 2 GB chip are exactly the same as the 1 GB chip
 on a normal XO).

 I did another test where I stubbed out the routine that writes the data to
 the USB key, just doing the NAND IO and scanning.  That took 2 minutes 40
 seconds (1 min 20 seconds for the first 1 GB).  So the USB I/O time was 4
 minutes 10 seconds.  The projected total time for a 1 GB machine would be
 5 minutes 30 seconds.

 That change, of course, doesn't really address the problem as reported -
 90 minute save times.  Something strange is happening; my results show
 that the basic structure of scan-nand is not the problem.

-- 
Ticket URL: <http://dev.laptop.org/ticket/8313#comment:4>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system