Super weird issue with surface write test and cardreader

[hdsentinelfan]

Hi

I spend a coupe hours testing out the cardreaders Transcend RDF8 and RDF9 using the following microSD cards:
- SanDisk Extreme Plus microSDXC 64GB UHS-I Class 10 U3
- Samsung PRO+ microSDXC 64GB UHS-I Class 10 U3

Now apart from the issues I've discovered with these cardreaders I found something super strange with the following setup:
1. Samsung PRO+ microSDXC 64GB UHS-I Class 10 U3
2. Transcend RDF9 cardreader
3. HD Sentinel 4.60 Pro surface test write

And here it is:

What-is-this.png (123.05 KiB) Viewed 19229 times

The dark green blocks write about 10-20 mb/s while the bright ones write with 90 mb/s.
I find the pattern super suspicious.
When I use the RDF8 cardreader I cannot reproduce the problem with the same card.
I already contacted Transcend support to find out more.

Would you guys have a clue what could be going on here ?

Thanks in advance,
Felix

[hdsentinel]

Thanks for the image.

I can confirm that this is normal, not rare situation when you perform testing of flash storage. This can happen with various card readers, memory cards, pendrives.

These devices may perform caching of disk writes (even if write cache disabled in Windows and also Hard Disk Sentinel disables write cache for the duration of the disk test). So when you actually see high write speed, the data is NOT actually written to the device, just after some time, when the cache is full. Then all data written once, so writing to the current block (when the data really written to previous blocks too) takes more time and that blocks seems slower. This results darker green color for that block.

Of course that one block is not slower than others - just when it is written, more blocks updated and this takes longer time.

If you re-run the test, maybe with different parameters, you may see different interesting "patterns".

For example if you select Disk menu -> Surface test and select the device, the test method - but before starting the test, click on the Configuration tab and select "Random test" or "Butterfly test" or "Sequential backward test" (and un-select the default Sequential test) to perform the test differently, you may see other patterns of darker green blocks, which will confirm that there is no real problem (no real slowness) with the previously shown darker areas.

[hdsentinelfan]

Awesome! Thanks a lot for your help

I am still super confused why this only happens with certain card readers and certain SD cards. It seems like the tech behind SD cards is more complicated than I thought.
The write pattern I pointed out above only happens with the Samsung Pro+ microSD card and only on the RDF9 cardreader. No problems with this cards on the RDF8 card.
As well as no problems with SanDisk Extreme Plus on the RDF9 and 8 reader.

Means that if there is caching of some kind then it it card related. But how would that work ? Would the reader identify the card model / ID and then decide how to write it ?

Meanwhile I got reply from Transcend. Friendly but not too helpful in a sense that they say that these are no problems and my stuff isnt reflecting normal usage which is kind of correct

[hdsentinel]

Yes, it is interesting that you find it with some specific combinations of cards, readers and so.
But yes, it is possible that different combinations may give raise some other, minor issues, for example a minor time-synchronization issue which results that the card, the reader (and/or even the USB 3.0 controller and its driver) may not 100% perform as fast as should.

Yes, I'm sure in such situations all manufacturers say everything is perfect - and in their viewpoint, it's true: the card can work perfectly, data can be read/written, maybe with a slight delay (especially with a particular card reader). I'm sure they can't (and will not want) to test all possible combinations of controllers, drivers, readers with their cards - and we still did not mention microSD-SD adapters, USB hubs, USB cable extenders. They should cause no such performance degradations or issues, but such weirdness can always happen

[hdsentinelfan]

Thank you very much for your reply and help

Meanwhile I spend some more time with my collection of USB 3.0 sticks and guess what: same issue depending on the usb stick type.

Then I looked into the benchmarks I did 1-2 years ago with these sticks and my last PC (Ivy Bridge, Z77).
I found that some of the sticks indeed always performed super bad on 4k writes.
But most were reading slower (4k) but writing faster with proper speed (4k, 2-3mb/s). So the controller (hardware, driver) is part of the problem.

As slow 4k write definitively affects my work with external storage media I really have to look closer at new purchases 4k performance.
A real pity that manufacturers don't really look into such "details".

Luckily I found a hint what the exact reason could be: http://www.computerbase.de/forum/showth ... st18222110 (german, sry )
The user refers to the wear leveling and mapping table being the reason for super slow writes even with new sticks.
Interesting enough this thread is about a new USB 3.0 stick Kingston HyperX Savage which shows exactly the same 4k write problem I am facing.

Thanks again for the feedback. I will live with the problem for now but will keep an eye on it and wont spend any money on products which ignore 4k write speeds