Re: XO-1.75 RAM [Devel Digest, Vol 62, Issue 25]

2011-04-14 Thread Yioryos Asprobounitis
 --
 
 Message: 4
 Date: Thu, 14 Apr 2011 10:36:10 +1000
 From: Sridhar Dhanapalan srid...@laptop.org.au
 Subject: Re: XO-1.75 RAM
 To: mi...@bga.com
 Cc: OLPC Devel devel@lists.laptop.org
 Message-ID: BANLkTi=_aasxf7++bmxoysaxvmxga4p...@mail.gmail.com
 Content-Type: text/plain; charset=ISO-8859-1
 
 On 14 April 2011 06:16, Mikus Grinbergs mi...@bga.com
 wrote:
  Might 1GB of RAM ?become a performance
 bottleneck?
 
  I myself am skeptical of efforts to use the OLPC where
 a desktop system
  might be more effective. ?I've run various large
 applications on an XO-1.5
  system, and have not myself experienced memory
 shortage.
 
  Unless the RAM chips on the XO-1.75 are socketed (and
 thus easily
  replaceable), it seems to me that it would be easier
 to add swap space
  (rather than RAM) to those systems used to run
 particularly memory-hungry
  applications (such as video editing).
 
 Flash storage is mush slower than spinning hard drives and
 wears out
 far more quickly. I don't think using flash for swap is a
 good idea.
 

For what it worths I have the same SDcard on the XO-1 with more than 1500 hours 
of swap use of a 256MB partition, with no problem so far.
On the XO-1.5 and its 1GB RAM a rarely see any swap use at all.

If anything I would rather see the extra cost towards a proper small SSD to 
replace SDcards, than extra RAM
___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Re: XO-1.75 RAM

2011-04-14 Thread James Cameron
On Thu, Apr 14, 2011 at 12:01:47AM -0700, Yioryos Asprobounitis wrote:
 For what it worths I have the same SDcard on the XO-1 with more than
 1500 hours of swap use of a 256MB partition, with no problem so far.

Cool.  What was the unit cost of the SD card, what size was it, and how
many writes were made over those 1500 hours to the SD card as a whole?

(... not just the swap partition, since endurance of the card relates to
total writes to the card).

-- 
James Cameron
http://quozl.linux.org.au/
___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Re: XO-1.75 RAM

2011-04-14 Thread Yioryos Asprobounitis
--- On Thu, 4/14/11, James Cameron qu...@laptop.org wrote:

 From: James Cameron qu...@laptop.org
 Subject: Re: XO-1.75 RAM
 To: Yioryos Asprobounitis mavrot...@yahoo.com
 Cc: devel@lists.laptop.org
 Date: Thursday, April 14, 2011, 3:33 AM
 On Thu, Apr 14, 2011 at 12:01:47AM
 -0700, Yioryos Asprobounitis wrote:
  For what it worths I have the same SDcard on the XO-1
 with more than
  1500 hours of swap use of a 256MB partition, with no
 problem so far.
 
 Cool.  What was the unit cost of the SD card, what
 size was it, 

That was/is a $15 Transcend 4GB class 6 card that not only can have 5 erase 
blocks open but also has an impressive random w/r speed (full flashbence data 
is attached)

 and how many writes were made over those 1500 hours to the SD card
 as a whole?
 
 (... not just the swap partition, since endurance of the
 card relates to
 total writes to the card).

Sorry no data on this, but the card is also holding an OS. Ubuntu 8.10 till 
recently and XOpup the last couple of weeks. 
Probably 1000 of the 1500 hours where run with this alternative OS, so I would 
think a lot of writes.
Two caveats though, the card spend half of its life with an ext2 file system 
and the other half with an ext3. 
My Ubuntu OS was configured with `vm.swappiness=0' (no swappiness settings for 
the OLPC builds in the NAND).

Transcend 4GB class6
Size: 3905536
/sys/devices/pci:00/:00:0c.0/mmc_host/mmc2/mmc2:8c10/name SU04G
/sys/block/mmcblk0/device/
name SDC
oemid 0x5356, 
manfid 0x1c, 
hwrev 0x1, 
fwrev 0x0, serial 
0x05593bca, 
scr 0235, 
csd 400e00325f591dcb7f800a40

bash-4.1# ./flashbench -a /dev/mmcblk0  --blocksize=1024
align 1073741824pre 604µs  on 669µs   post 576µs diff 
79.1µs
align 536870912 pre 586µs  on 658µs   post 568µs diff 81.3µs
align 268435456 pre 579µs  on 658µs   post 567µs diff 85.1µs
align 134217728 pre 585µs  on 655µs   post 563µs diff 81.3µs
align 67108864  pre 582µs  on 661µs   post 554µs diff 92.6µs
align 33554432  pre 582µs  on 653µs   post 563µs diff 80.6µs
align 16777216  pre 587µs  on 660µs   post 562µs diff 85.5µs
align 8388608   pre 584µs  on 669µs   post 566µs diff 93.7µs
align 4194304   pre 589µs  on 664µs   post 567µs diff 86.6µs
align 2097152   pre 585µs  on 659µs   post 569µs diff 81.3µs

align 1048576   pre 588µs  on 643µs   post 594µs diff 52.2µs
align 524288pre 588µs  on 640µs   post 584µs diff 54.3µs
align 262144pre 580µs  on 638µs   post 587µs diff 54.4µs
align 131072pre 584µs  on 635µs   post 586µs diff 49.9µs
align 65536 pre 594µs  on 635µs   post 588µs diff 43.8µs
align 32768 pre 571µs  on 614µs   post 563µs diff 46.9µs
align 16384 pre 574µs  on 609µs   post 565µs diff 39.6µs
align 8192  pre 572µs  on 614µs   post 562µs diff 46.9µs
align 4096  pre 573µs  on 610µs   post 565µs diff 41.3µs

align 2048  pre 560µs  on 564µs   post 558µs diff 5.55µs


bash-4.1# ./flashbench -O --erasesize=$[2 * 1024 * 1024] --blocksize=$[16 * 
1024] /dev/mmcblk0 --open-au-nr=4
2MiB9.38M/s 
1MiB8.67M/s 
512KiB  8.88M/s 
256KiB  8.49M/s 
128KiB  8.63M/s 
64KiB   9.83M/s 
32KiB   10.6M/s 
16KiB   8.59M/s 
bash-4.1# ./flashbench -O --erasesize=$[2 * 1024 * 1024] --blocksize=$[16 * 
1024] /dev/mmcblk0 --open-au-nr=5
2MiB9.96M/s 
1MiB9.28M/s 
512KiB  9.5M/s  
256KiB  9.06M/s 
128KiB  9.22M/s 
64KiB   10.2M/s 
32KiB   10.5M/s 
16KiB   8.7M/s 
bash-4.1# ./flashbench -O --erasesize=$[2 * 1024 * 1024] --blocksize=$[16 * 
1024] /dev/mmcblk0 --open-au-nr=6
2MiB6.65M/s 
1MiB5.93M/s 
512KiB  2.45M/s 
256KiB  1.35M/s 
128KiB  720K/s  
64KiB   381K/s  
32KiB   194K/s  
16KiB   97.6K/s

bash-4.1# ./flashbench -O --erasesize=$[2 * 1024 * 1024] --blocksize=$[16 * 
1024] /dev/mmcblk0  --random --open-au-nr=4
2MiB5.96M/s 
1MiB5.55M/s 
512KiB  5.1M/s  
256KiB  4.91M/s 
128KiB  4.8M/s  
64KiB   4.75M/s 
32KiB   4.09M/s 
16KiB   3.3M/s  
bash-4.1# ./flashbench -O --erasesize=$[2 * 1024 * 1024] --blocksize=$[16 * 
1024] /dev/mmcblk0  --random --open-au-nr=5
2MiB6.05M/s 
1MiB5.61M/s 
512KiB  5.12M/s 
256KiB  4.91M/s 
128KiB  4.8M/s  
64KiB   4.7M/s  
32KiB   4.11M/s 
16KiB   3.28M/s 
bash-4.1# ./flashbench -O --erasesize=$[2 * 1024 * 1024] --blocksize=$[4 * 
1024] /dev/mmcblk0  --random --open-au-nr=6
2MiB7.97M/s 
1MiB4.28M/s 
512KiB  2.52M/s 
256KiB  1.36M/s 
128KiB  722K/s  
64KiB   378K/s  
32KiB   193K/s
16KiB   97.5K/s

bash-4.1# ./flashbench --erasesize=$[2 * 1024 * 1024] --findfat --random 
/dev/mmcblk0
2MiB5M/s 7.62M/s  7.14M/s  12.6M/s  12.5M/s  12.3M/s  
1MiB4.78M/s  4.85M/s  4.82M/s  3.46M/s  3.47M/s  3.52M/s  
512KiB  3.52M/s  3.54M/s  3.53M/s  3.53M/s  3.53M/s

Re: XO-1.75 RAM

2011-04-14 Thread James Cameron
On Thu, Apr 14, 2011 at 01:21:23AM -0700, Yioryos Asprobounitis wrote:
 That was/is a $15 Transcend 4GB class 6 card that not only can have 5
 erase blocks open but also has an impressive random w/r speed (full
 flashbence data is attached)

Okay, thanks.  I don't know a deployment willing to fork out an extra
$15 per laptop for performance ... but others in my team might know.

Sounds like a very nice card if one can afford it, and if the current
batch persists in the same endurance and performance.

-- 
James Cameron
http://quozl.linux.org.au/
___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Re: XO-1.75 RAM

2011-04-13 Thread Mikus Grinbergs
 Might 1GB of RAM  become a performance bottleneck?

I myself am skeptical of efforts to use the OLPC where a desktop system 
might be more effective.  I've run various large applications on an 
XO-1.5 system, and have not myself experienced memory shortage.

Unless the RAM chips on the XO-1.75 are socketed (and thus easily 
replaceable), it seems to me that it would be easier to add swap space 
(rather than RAM) to those systems used to run particularly 
memory-hungry applications (such as video editing).

mikus

___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Re: XO-1.75 RAM

2011-04-13 Thread Sridhar Dhanapalan
On 14 April 2011 06:16, Mikus Grinbergs mi...@bga.com wrote:
 Might 1GB of RAM  become a performance bottleneck?

 I myself am skeptical of efforts to use the OLPC where a desktop system
 might be more effective.  I've run various large applications on an XO-1.5
 system, and have not myself experienced memory shortage.

 Unless the RAM chips on the XO-1.75 are socketed (and thus easily
 replaceable), it seems to me that it would be easier to add swap space
 (rather than RAM) to those systems used to run particularly memory-hungry
 applications (such as video editing).

Flash storage is mush slower than spinning hard drives and wears out
far more quickly. I don't think using flash for swap is a good idea.

More RAM can be used for write caching, which can greatly improve
performance (depending on the load).

Sridhar
___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Re: XO-1.75 RAM

2011-04-13 Thread James Cameron
On 14 April 2011 06:16, Mikus Grinbergs mi...@bga.com wrote:
 Unless the RAM chips on the XO-1.75 are socketed (and thus easily
 replaceable), 

This seems unlikely.  Socketing adds cost up front, and low cost is a
primary goal.  Also the demand for upgrade would be relatively low.

 ... particularly memory-hungry applications (such as video editing).

Video editing requires low latency storage rather than large amounts of
memory.  The files, being often larger than memory, cause the cache to be
filled with single-use data, which is a waste.  You get the same cache
effect just by copying the files.

-- 
James Cameron
http://quozl.linux.org.au/
___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel