Comparing UBI volumes to generate patch for upgrades

Comparing UBI volumes to generate patch for upgrades

[Subodh Nijsure]

Hi,

I am are working on a embedded industrial product that uses UBIFS, but 
we have limitation on amount of data this product can download per year.

So we are trying to figure out how to optimise size of upgrade package.

one option is  use VCDIFF/bsdif or variation there-of and generate a 
binary diff package.

option 2 is write a tool that compares UBI volume layout for revision 1 
and UBI volume layout for revision 2 and then send that difference, 
along the lines of what is outlined in the paper @ 
http://www.icmu.org/icmu2010/Papers/3-1.pdf

We are just getting started writing a tool to compare UBI volumes, but I 
want to send out querry to UBI experts to see if there are some 
fundamental issues with UBI that would make comparing UBI volumes and 
applying such a patch would be a non-starter.

If we do come up with a "decent" enough tool that can compare UBI 
volumes will certainly share it.

-Sam

Re: Comparing UBI volumes to generate patch for upgrades

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 968 bytes --]

On Wed, 2012-02-01 at 10:31 -0800, Subodh Nijsure wrote:
> one option is  use VCDIFF/bsdif or variation there-of and generate a 
> binary diff package.
> 
> option 2 is write a tool that compares UBI volume layout for revision 1 
> and UBI volume layout for revision 2 and then send that difference, 
> along the lines of what is outlined in the paper @ 
> http://www.icmu.org/icmu2010/Papers/3-1.pdf

I guess you can do this for R/O volumes, but not R/W.

Depending on how much extra flash you have, you may do the actual update
in a power-off-tolerant manner or not.

> We are just getting started writing a tool to compare UBI volumes, but I 
> want to send out querry to UBI experts to see if there are some 
> fundamental issues with UBI that would make comparing UBI volumes and 
> applying such a patch would be a non-starter.

I assume you are not planning to do this stuff in kernel space, right?

-- 
Best Regards,
Artem Bityutskiy


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: Comparing UBI volumes to generate patch for upgrades

[Subodh Nijsure]

On 02/03/2012 01:18 AM, Artem Bityutskiy wrote:
> On Wed, 2012-02-01 at 10:31 -0800, Subodh Nijsure wrote:
>> one option is  use VCDIFF/bsdif or variation there-of and generate a
>> binary diff package.
>>
>> option 2 is write a tool that compares UBI volume layout for revision 1
>> and UBI volume layout for revision 2 and then send that difference,
>> along the lines of what is outlined in the paper @
>> http://www.icmu.org/icmu2010/Papers/3-1.pdf
> I guess you can do this for R/O volumes, but not R/W.
Yes we would do this 'upgrade' on read-only volume.
>
> Depending on how much extra flash you have, you may do the actual update
> in a power-off-tolerant manner or not.
Actually we do have to do this upgrade in power-off-tolerant manner, but 
hardware does give us few seconds of "battery backed" power, and 
indicates to software that shutdown is imminent.
>> We are just getting started writing a tool to compare UBI volumes, but I
>> want to send out querry to UBI experts to see if there are some
>> fundamental issues with UBI that would make comparing UBI volumes and
>> applying such a patch would be a non-starter.
> I assume you are not planning to do this stuff in kernel space, right?
Not planning to this in kernel space.