Transfer rate of MS PRO, MS, & CF

[jlbunting]

I have done some checking on different web sites and I wanted to make sure I understand this correctly.

The transfer Rate (most realistic) for:

MS: (128)
Access Write Speed Max 1.5 MB/second
Access Read Speed Max 2.45 MB/second

MS Pro: ( All sizes)
transfer rate of up to 160 megabits per second (or 20MB/sec) and a write speed of over 15 Mbps

CF:
Capacity/ Read Speed / Write Speed / Average MB/sec / X Rate
128MB 5.4 2.2 4.5 30X
1GB 5.2 3.2 4.5 30X


IF I am Wrong please some one let me know, If I am right I would like to know this as well.
Thanks.

[jlbunting]

I fixed the CF table> hopefuly the table will be formated correctly this time>

CF:
Capacity/ Read Speed / Write Speed / Average MB/sec / X Rate
128MB / 5.4 / 2.2 / 4.5 / 30X
1GB / 5.2 / 3.2 / 4.5 / 30X

[n2ifp]

It's not a realistic comparison. The MS Pro probably won't be much faster that the MS or CF. In a native MS Pro device yes, but in those devices needing the driver no, source Sony.

[YTTAN]

Not all CF had the same speed. Apacer is the fastest one(2.4-5mb). Even faster than sandisk / lexar (about 1.5-2.5mb/sec only) and other branded too!

[jlbunting]

I only wonder which would be the best choice> when you take into account Speed, Cost and Size? CF,MS< or MS pro.

[rhart00]

I think I remember seeing an app on palmgear that measures read/write speeds of VFS memory but I can't remember the name. If it worked with the CF drivers on would give us a nice clie specific comparision of read/write times on the different types of memory.

[YTTAN]

jlbunting,

The best choice when you take into account Speed, Cost between the CF,MS or MS pro is====* Apacer CF *of course in Sony clieworld.

Ms Pro is faster in other digi cam at 160mbps--20MB/sec only, not in clie.(20mbps--2.5MB/sec).

Talking about sizes CF is bigger but that it's ok as it is still inside my NZ90 or NX70 which just like ms or ms pro are inside the clie too.

[n2ifp]

Lately, the CF is much faster than the MS in the Clie :)!

Before doing anything, wait until we see how the CF Driver turns out.

[greg elmassian]

I have both normal ms, and a 256k pro... backup program ran 2.5 minutes with normal stick, and 1.5 minutes with pro...

Greg

[YTTAN]

May I ask how many MB you had transfer for 1.5min for MS Pro?

[greg elmassian]

I think I had 5 megs of free ram, so that would be about 10 megs? I can do some tests this weekend to give better numbers, all I am interested in is how much faster it was, and every little bit helps with backups (that means I do them more often),.... by the way, that was backupbuddy vfs, not the stupid internal backup program.

Greg

[YTTAN]

ok thanks.

[n2ifp]

Originally posted by greg elmassian
I have both normal ms, and a 256k pro... backup program ran 2.5 minutes with normal stick, and 1.5 minutes with pro...

Greg

Greg,

You don't know yet how the MS Pro driver will affect speed.

[Drachen]

I just checked ran VSMark against my two cards on my NZ90, and the MS-pro is faster than the garden variety MS. My Sony 512MB MS-pro was rated 212 and my Sandisk 128MB stick got a 149.

Here's the program I used if you're curious:
http://www.freewarepalm.com/utilities/vfsmark.shtml

[greg elmassian]

Larry, you are right, but I would suspect the "driver" will not affect the speed within the nz-90, but only enable the msimport product to work. I doubt seriously that any other change needs to be done, because the pro is acessed just fine by the nz.

Therefore my assumption is that the basic data transfer is ok, and they would have no reason to "touch" it. Looks as though the MSImport and other transfer programs just don't use the existing internal driver properly.

Greg

[greg elmassian]

vfsmark:

ms pro 256: 216
sony 128: 149
sandisk 128: 140

[YTTAN]

Good MS Pro is faster by 2 time without patch!

What about CF speed especially from Apacer?

[greg elmassian]

YTTAN, normally I can decode what you are saying?

I don't understand your comment... are you trying to make a statement that only "good" ms pro's will be twice as fast as a normal ms? well, it's not true in our nz's....

and until someone in the beta does vfs on their cf, we won't know... i suppose that is what you are asking?

Suggestion:

"I thought a MS PRO would be twice as fast" But your measurements show it is not.

"Will the MS PRO patch make it faster?" (my opinion is no, as stated before)

Will someone in the beta driver program please test a CF with vfsmark?

Regards,

Greg

(no offense, but spending a little more time on the message would make it a lot more understandable, and not ambiguous)

[CliePet]

re: VFSMark stats

If posting VFSMark statistics, please include more than just the VFSMark total.
IMHO the total is not that interesting.

----

In particular please include the "File Write" and "File Read" percentages (relative to the comparison model). These are the most interesting (especially for large file reads/writes).

The write will be significantly slower than the read (and may fluctuate depending on the stick)

NOTE: all flash memory storage devices have an uneven write time (it varies with the chips, and even the blocks within the chips).

I suspect the MS Pro read time to be approximately the same speed as a normal memory stick (constrained by the serial bus). If the write speed is faster, it is probably because they are using newer/better flash ROM chips.

[Drachen]

Here are the VFSMark trials I did this morning:

Sony MS-pro 512:
VFSMark Results

File Create: 119%
File Delete: 116%
File Write: 20%
File Read: 236%
File Seek: 251%
DB Export: 103%
DB Import: 377%
Record Access: 344%
Resource Access: 322%

VFSMark: 209


Sandisk MS 128:
VFSMark Results

File Create: 112%
File Delete: 63%
File Write: 27%
File Read: 144%
File Seek: 190%
DB Export: 80%
DB Import: 242%
Record Access: 236%
Resource Access: 215%

VFSMark: 145


The results don't seem to bear out your conclusions about the serial bus, since the pro file read score is 60% faster than the standard MS. It's faster across the board except for the File Write score, and both File Write scores are much lower than the baseline m500 + 16MB SD card.

[YTTAN]

greg elmassian,

The following is actually meaning:
___________________________________________________
That is a Good New!! You data had showed that MS Pro is faster than normal MS by 2 time without any patch from Sony!
___________________________________________________

Sorry for your misunderstanding in this case.

[YTTAN]

My Sony MS 128 with NZ90/g:
VFSMark Results

File Create: 112%
File Delete: 63%
File Write: 27%
File Read: 138%
File Seek: 188%
DB Export: 79%
DB Import: 242%
Record Access: 232%
Resource Access: 213%

VFSMark: 143

How accurate we can trust on this vfsmark?

When I tried it for 2nd time, I got different score!!!

[greg elmassian]

i think the difference is slight... and to be expected on benchmarks, rounding error, and the file read/write could be different if the files on the MS were changed... but they are close enough to be reasonable...

Greg

[CliePet]

> How accurate we can trust on this vfsmark?

Ok, VFSMark is a poor program for measuring raw I/O performance to the storage device.

I too get different numbers with different runs (same stick, same CLIE). Also the numbers are very different when run on an NR70 with the same stick (same stick, slower CPU). Ie. you would expect the raw i/o transfer rate to be relatively independent of CPU speed.

----
Kudos for Kopsis Inc for releasing the source code under GPL, so you can see what they are actually measuring.

It measures operating system overhead and other things and tries to reduce many statistics to one number (the problem with many 'benchmarks')

eg: the read includes 1000 calls for small reads (16B), and 100 calls for large reads (16KB), then adds the times together. Also the read/write are unbalanced (spending much more time doing writes than reads)

I'll whip up a quick raw read/write test program for people to run instead, and see if that makes more sense.

[CliePet]

Here's a simple program that just does raw reads and raw writes to the memory stick. File open/close overhead is not counted

http://aibohack.com/clie/Time_RawVFS.prc

A memory stick with 1MB of free space is needed

Takes about 1 minute to run [10 MB of raw read, 10MB raw write]


===============================

My numbers with Time_RawVFS

CLIE NZ90 [no driver updates], standard 64MB Memory stick:

Read: 636 KB/sec
Write: 279 KB/sec

[ numbers divided by 1000 for KB/sec ratings]

Subsequent runs get some variation in timing. Results vary a little with the stick tested.
For some reason an MG stick is a little slower (but could be typical variations due to chip speed)


CLIE NR70, same memory stick
Read: 701 KB/sec
Write: 307 KB/sec

ie. confirms the hypothesis the raw stick speed doesn't change much with CPU speed, unlike the VFSMark numbers,
and in fact is faster on the NR70 (perhaps because of no 68K/ARM overhead??)

-------

If someone with an MS Pro stick would run this program, that would be appreciated.

If someone on the CF driver beta could post similar results, that would be appreciated too, assuming it doesn't violate your NDA.

It should run the test for all VFS volumes in order discovered.
Be sure your Memory Stick (or CF card) is inserted before you run the test.

I will run comparison tests for PocketPC later (Memstick, CF etc). The CF->MS adapter appears to give similar numbers to the above (ie. still constrained by the serial bus of the memory stick)


Thanks

[Drachen]

Cool little program. Here are my stats on an NZ90 with my Sony MS-pro 512:
read = 987 KB/s
write = 205 KB/s

Here's my Sandisk 128MB MS for comparison:
read = 589 KB/s
write = 283 KB/s

That's about the same results, percentage-wise, as the VFSMark File Read and File Write results I had earlier. Reading from the MS-pro is about 60% faster than the MS and writing is about 30% slower.

[n2ifp]

Originally posted by CliePet
re: VFSMark stats

If posting VFSMark statistics, please include more than just the VFSMark total.
IMHO the total is not that interesting.

----

In particular please include the "File Write" and "File Read" percentages (relative to the comparison model). These are the most interesting (especially for large file reads/writes).

The write will be significantly slower than the read (and may fluctuate depending on the stick)

NOTE: all flash memory storage devices have an uneven write time (it varies with the chips, and even the blocks within the chips).

I suspect the MS Pro read time to be approximately the same speed as a normal memory stick (constrained by the serial bus). If the write speed is faster, it is probably because they are using newer/better flash ROM chips.

That is basically what Sony has said, the speed will be about the same. On native devices, then the MS Pro should be faster. I suspect that the CF memory will give the best results for the Clie. It's still too early to make any real world tests. One fellow did a comparison on all three with the CF coming out on top. I don't remember what was used to benchmark it.

[mvfrancisco]

Would the clie model affect read/write speed? Has anyone tried it on a TG-50?

[CliePet]

Drachen wrote:
> with my Sony MS-pro 512:
> read = 987 KB/s
> write = 205 KB/s

That's still strange.

I can understand the read being faster (the low level serial bus will go up to 1.5 MB / sec)

One of the advertised features of the MS Pro sticks is a minumum write speed of 15Mbps (>1.8 MBps) for cameras and the like. Unless they are doing something stupid, the writing to flash chip should not be a bottleneck.

I'm not sure how much is file system overhead v. slot driver overhead v. serial bus overhead v. actual flash speed. Also Magic Gate may be getting in the way.

BTW: My PocketPC CF card maxes out at around 400 KB/sec raw write speed using a similar test - so even it has far more overhead than necessary.

------
mvfrancisco wrote:
> Would the clie model affect read/write speed

Yes, especially if you use the VFSMark program to measure overall memory stick performance.

In my case, my older NV70 is faster at raw read and write, but has a significantly lower VFSMark number.

[greg elmassian]

all that really matters is how fast it runs applications in your pda... for me, the backup is the thing, and the ms pro backed up faster than the normal ms... so i'm happy..

i'm waiting for my microdrive to be supported (i.e. cf driver), the microdrive, while using more power, writes faster than most cf...

greg

[n2ifp]

Originally posted by greg elmassian
all that really matters is how fast it runs applications in your pda... for me, the backup is the thing, and the ms pro backed up faster than the normal ms... so i'm happy..

i'm waiting for my microdrive to be supported (i.e. cf driver), the microdrive, while using more power, writes faster than most cf...

greg

The microdrive does work, that's all I remember.

[greg elmassian]

from my testing on the trgpro, and the handera 330, the write speed on the microdrive was noticably faster than a cf ... and i have a few cf's!!!

i'll make you a gentlemen's bet that it is faster on write speed (though some wouldn't care due to power consumption)

greg

(too bad you aren't in san diego, we could check it out)

73 de N6RGZ

[somas1]

Originally posted by greg elmassian
all that really matters is how fast it runs applications in your pda... for me, the backup is the thing, and the ms pro backed up faster than the normal ms... so i'm happy..

i'm waiting for my microdrive to be supported (i.e. cf driver), the microdrive, while using more power, writes faster than most cf...

greg

What is as important to me as how fast applications on the card run is how quickly I can get files onto the card.&nbsp; If I save $150 by buying a cf card but then find out that I can only get 1 or 2 megabytes per minute onto the card then it is almost useless since it would take 2 hours to get 120 MB of movies and music onte the card.&nbsp;

[n2ifp]

Originally posted by greg elmassian
from my testing on the trgpro, and the handera 330, the write speed on the microdrive was noticably faster than a cf ... and i have a few cf's!!!

i'll make you a gentlemen's bet that it is faster on write speed (though some wouldn't care due to power consumption)

greg

(too bad you aren't in san diego, we could check it out)

73 de N6RGZ

Greg,

No argument from me, it just shows to be very promising!

Best regards!

&nbsp;

[greg elmassian]

yep, keeping my fingers crossed!

bisides, i'm a detail guy, and that microdrive fills that slot perfectly!

Greg

[YTTAN]

Here's my Sony 128MB MS for comparison:
read = 580 KB/s
write = 283 KB/s

I don't understand why Sony ms is slower than sandisk.

[YTTAN]

Here's my Sony 128MB MS 2nd reading:
read = 914 KB/s
write = 283 KB/s

??? not accurate again???

[n2ifp]

Very perceptive of you :)!

[CliePet]

somas1 wrote:
> What is as important to me as how fast applications on the card run is how quickly I can get files onto the card.

Typically people use a PC CF adapter (USB -> CF) to store data to a CF card (in the case of movies at least). and then play it on the PDA.

USB adapters are still pretty slow, not because of USB (12MB/sec), but mostly because of the flash-write speed. Typically they are faster than PDA access for a number of reasons.

The actual rate will depend on the CF card, but also the CF adapter you use.

I rough estimate I like to use is to assume 50MB/minute for bulk copies (your mileage will vary).

-----
YTTAN wrote:
> read = 914 KB/s
> ??? not accurate again???

You will see some variation in the numbers, from stick to stick.
Subsequent runs on the same stick should be in the same ball-park.

Please run it a few more times. The 914 KB/s number looks way too high (perhaps a measurement problem)

AFAIK there is no real buffering of the stick data, but my quick-and-dirty sampling program suffers some of the same clock roundoff problems as VFSMark

[greg elmassian]

or the faster result is that there is a cache...

for all of it's weaknesses, VFSMark has been used on a lot of PDA's and is a good relative measurement... there is a lot of data available...

once you are in the ballpark (VFSMark shows something is faster or slower)... then I would suggest you time the execution of your favorite programs/operations...

You can then decide if the extra performance of a MS PRO is worth it to you or not.

Greg

[legion]

For reference vfsmark rated my 128mb SD card in my Palm 505 as 63

[dark79]

Here's my results with VFSMark:

Sandisk 128 MS: Apacer Photo Steno Pro 512MB CF:

File Create: 117% File Create: 120%
File Delete: 65% File Delete: 60%
File Write: 28% File Write: 38%
File Read: 139% File Read: 224%
File Seek: 184% File Seek: 193%
DB Export: 83% DB Export: 89%
DB Import: 239% DB Import: 190%
Record Access: 264% Record Access: 266%
Resource Access: 238% Resource Access: 225%

VFSMark: 150 VFSMark: 156


Results with Time Raw-VFS:

Sandisk 128 MS: Apacer Steno Photo Pro 512MB CF:

read = 574 kb/s read = 1,047 kb/s
write = 283 kb/s write = 243 kb/s

[kdn102]

No way....pocketPC running a windows based product having more overhead than necessary???&nbsp; I can't believe it!

That's a little sarcasm.

And that last statement had even more sarcasm.

I just decided I have jet lag......so I better stop while I'm behind.

Originally posted by CliePet

BTW: My PocketPC CF card maxes out at around 400 KB/sec raw write speed using a similar test - so even it has far more overhead than necessary.

[djr]

Hi,

I know this thread is four months old, but I'm struggling with the results of CLIE Pets' Time_RawVFS program.
He states here (http://www.cliesource.com/forums/showthread.php?s=&threadid=10401&perpage=15&pagenumber=2):

Subsequent runs get some variation in timing. Results vary a little with the stick tested.

What's causing these variations? Is there some caching involved? The following tests were run on a Palm Tungsten|T with a 256MB SD Lexar card, with five minutes of doing nothing between each run (no alterations on Palm or card)

Read Write
Test 1
1KB 396 21
4KB 494 69
16KB 529 199
Test 2
1KB 397 25
4KB 494 109
16KB 529 245
Test 3
1KB 396 32
4KB 494 109
16KB 529 253

[djr]

Sorry, posted as guest again, so can't edit :(

As you might have noticed the read numbers are pretty constant, but the write numbers increase with every run. That's why I thought about the caching, but that's pretty odd, because you want to cache reading and not writing.

- jp

[patrickl]

I ran some tests on my Clie NX70 with a Dane Elec CF card and got wildly varying results too. In my case only the read speeds varied a lot. The write speeds had less variation.

CF 256MB Dane Elec (Sony CF drivers)
Read Write
Test 1
1kB 500 23
4kB 782 94
16kB 1221 270
Test2
1kB 501 23
4kB 780 94
16kB 866 293
Test3
1kB 706 22
4kB 775 94
16kB 1221 270
Test 4
1kB 500 23
4kB 1159 88
16kB 871 293
Test 5
1kB 499 23
4kB 777 94
16kB 1217 271


So, 4kB read varies between 777 and 1159, 16kB read varies between 866 and 1221.

Interestingly the write speed seems reversly coupled to the read speeds. When the read speed makes a weird spike the write speed drops and vice versa.

With my 512MB MS Pro I see no variation between tests. That might just have been luck, but I tried 10 runs and didnt get a difference above 1%.

[CliePet]

re: timing variations

I am doing no caching, and reading/writing big enough blocks that caching shouldn't matter. However when using the EruWare caching option, my tests did speed up, so there is something going on.

The basic timing logic has some roundoff errors (code from VFSMark adapted for my purposes). I'll look into it, because that may explain your read/write numbers.

----
In general:
The read times should be in the same ballpark.

The write times can vary - because programming flash memory can take a uncertain amount of time [because of the electron tunnelling technique used]

[patrickl]

I did some more tests and I seem to get two possible timing sets. The overall results for me consist of combinations of these sets:
Slow read set Fast read set
Read Write Read Write
1kB 505 23 1kB 730 22
4kB 777 94 4kB 1186 88
16kB 868 293 16kB 1232 271

Sometimes I did get many results the same though. For instance the last 10 tests all gave me the fastest results.

BTW I'm using the Sony CF drivers (don't have the EruWare ones).

I also tried VFSMark and I also got some variations in read (between 178% and 184%) and write speeds (between 36% and 50%) with it. Obviously that's not as big a difference as with Time_RawVFS, but that might be because the results in the VFSMark Read test are averaged over several tests.

Weird how with VFSMark only my write speeds seemed to vary significantly (more than 5%) and with Time_RawVFS only the read speeds vary enough to be noticed.

Also, I was kinda surprised that Lightspeed made quite an impact. With a 512MB MS Pro I got:
Lightspeed on Lightspeed off
Read Write Read Write
1kB 923 12 1kB 512 12
4kB 1542 51 4kB 824 50
16kB 1863 206 16kB 983 200

Read speed almost doubled! (I tried several runs, but again no variation between runs in the speeds for the MS Pro)


BTW where is the PocketPC version of Time_RawVFS?

[nx70]

Sorry for being a bit to late..
Did I get this rifght, the CF- cards are faster than the MS sticks? I have as well tried Time Raw and it gives me up to ten times faster acces to cf card than to MS. Is there a problem or is it really that fast? How come?