| 1 |
On Sun, Jun 17, 2012 at 7:58 PM, Alan McKinnon <alan.mckinnon@×××××.com> wrote: |
| 2 |
> On Sun, 17 Jun 2012 19:27:36 -0400 |
| 3 |
> Michael Mol <mikemol@×××××.com> wrote: |
| 4 |
> |
| 5 |
>> So, I'm using dd to write the Gentoo LiveDVD ISO directly to an SDHC |
| 6 |
>> card. I was wondering why the transfer rate had slowed from 20MB/s to |
| 7 |
>> 6MB/s, so I ran 'sudo tail -f /var/log/messages' to see if there were |
| 8 |
>> I/O errors slowing things down. |
| 9 |
>> |
| 10 |
>> What I found was a flood-stream of message blocks that look like this: |
| 11 |
|
| 12 |
[snip] |
| 13 |
|
| 14 |
>> ... Any idea what kernel configuration flag I may have enabled to |
| 15 |
>> cause these to be continually generated? My first guess would be |
| 16 |
>> something like 'kernel lock debugging'...if that's what I'm seeing |
| 17 |
>> here, where would I go to file a bug report against the usb-storage |
| 18 |
>> subsystem? |
| 19 |
>> |
| 20 |
> |
| 21 |
> 6M is what you should expect from a good Class 6 SDHC card, and those |
| 22 |
> are quite rare in consumer shops (few carry better than Class 4 in my |
| 23 |
> experience). |
| 24 |
|
| 25 |
Yeah, for a given class N, N MB/s is minimum certified transfer rate. |
| 26 |
Though they probably don't take cover block reflashing in that... |
| 27 |
|
| 28 |
> Writes start very fast (it's going into the kernel buffer) |
| 29 |
> then gradually slow down to the card's actual speed, which is what you |
| 30 |
> appear to be seeing. |
| 31 |
|
| 32 |
That kernel buffer would have been a gigabyte or two in size, and I |
| 33 |
don't remember noticing that while watching in htop. |
| 34 |
|
| 35 |
> |
| 36 |
> 20M write speed would be a Class 20, which doesn't exist yet :-) |
| 37 |
|
| 38 |
Well, more to the point there's no certification level for it. Devices |
| 39 |
can exceed their certification level. |
| 40 |
|
| 41 |
> But what is your actual query? About the write speed? |
| 42 |
|
| 43 |
No, not really. TBH, I wasn't expecting to be able to consistently |
| 44 |
write quickly to the beginning of the card, but slowly at the end; I |
| 45 |
wanted to make sure the card wasn't going bad given the abruptness of |
| 46 |
the change. Actually, the more I think about it, I think it happened |
| 47 |
at the 2GB boundary, where some funky pin logic changes. (And where SD |
| 48 |
becomes SDHC). So I think that's where the performance shifted. Maybe |
| 49 |
I'll invest in a bunch of fast 2G cards, if it means the performance |
| 50 |
is consistent, and I'd still fit just under 200 shots. |
| 51 |
|
| 52 |
> |
| 53 |
> Or about the presence of the messages? |
| 54 |
|
| 55 |
Yeah, it's the messages I was curious about. |
| 56 |
|
| 57 |
-- |
| 58 |
:wq |
Archives