Suggestion needed to set proper journal size

Suggestion needed to set proper journal size

[Subodh Nijsure]

Hello,

I know general implications of changing journal size on UBIFS -- bigger 
the journal,
longer it may take to replay the journal, increasing time UBIFS takes to 
mount. Smaller
journal means more re-writes to flash.

We created 2 partitions 100MB and 45MB and we see that by default 
mkfs.ubifs is reserving 8MB for journal.

UBIFS: file system size:   101834752 bytes (99448 KiB, 97 MiB, 802 LEBs)
UBIFS: journal size:         9023488 bytes (8812 KiB, 8 MiB, 72 LEBs)

UBIFS: file system size:     47996928 bytes (46872 KiB, 45 MiB, 378 LEBs)
UBIFS: journal size:          9023488 bytes (8812 KiB, 8 MiB, 72 LEBs)

I know I can specify size of journal with -j option.

Should the journal size be scaled in some proportion of size of UBIFS 
partition?

Should the journal size depend on how often you expect file-system to be 
updated?

Has anyone done experiments with setting really small (1MB/2MB) journal 
size?
What side effects is one likely to encounter if the journal size is too 
small.

fwiw in my use case, file-system doesn't get updated that often, perhaps 
2K data in about 4 files once a week.

-Subodh

Re: Suggestion needed to set proper journal size

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 1580 bytes --]

On Fri, 2012-04-20 at 15:59 -0700, Subodh Nijsure wrote:
> Hello,
> 
> I know general implications of changing journal size on UBIFS -- bigger 
> the journal,
> longer it may take to replay the journal, increasing time UBIFS takes to 
> mount. Smaller
> journal means more re-writes to flash.
> 
> We created 2 partitions 100MB and 45MB and we see that by default 
> mkfs.ubifs is reserving 8MB for journal.
> 
> UBIFS: file system size:   101834752 bytes (99448 KiB, 97 MiB, 802 LEBs)
> UBIFS: journal size:         9023488 bytes (8812 KiB, 8 MiB, 72 LEBs)
> 
> UBIFS: file system size:     47996928 bytes (46872 KiB, 45 MiB, 378 LEBs)
> UBIFS: journal size:          9023488 bytes (8812 KiB, 8 MiB, 72 LEBs)
> 
> I know I can specify size of journal with -j option.
> 
> Should the journal size be scaled in some proportion of size of UBIFS 
> partition?

It does grow, but not infinitely. We have some heuristics in mkfs.ubifs,
but those were mostly taken out of thin air. I do not remember us doing
a lot of experimenting with that.

> Should the journal size depend on how often you expect file-system to be 
> updated?
> 
> Has anyone done experiments with setting really small (1MB/2MB) journal 
> size?
> What side effects is one likely to encounter if the journal size is too 
> small.
> 
> fwiw in my use case, file-system doesn't get updated that often, perhaps 
> 2K data in about 4 files once a week.

I think you should do your own experiments. If you share the results -
will be nice.

-- 
Best Regards,
Artem Bityutskiy

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

Re: Suggestion needed to set proper journal size

[Subodh Nijsure]

On 04/27/2012 07:05 AM, Artem Bityutskiy wrote:
> On Fri, 2012-04-20 at 15:59 -0700, Subodh Nijsure wrote:
>> Hello,
>>
>> I know general implications of changing journal size on UBIFS -- bigger
>> the journal,
>> longer it may take to replay the journal, increasing time UBIFS takes to
>> mount. Smaller
>> journal means more re-writes to flash.
>>
>> We created 2 partitions 100MB and 45MB and we see that by default
>> mkfs.ubifs is reserving 8MB for journal.
>>
>> UBIFS: file system size:   101834752 bytes (99448 KiB, 97 MiB, 802 LEBs)
>> UBIFS: journal size:         9023488 bytes (8812 KiB, 8 MiB, 72 LEBs)
>>
>> UBIFS: file system size:     47996928 bytes (46872 KiB, 45 MiB, 378 LEBs)
>> UBIFS: journal size:          9023488 bytes (8812 KiB, 8 MiB, 72 LEBs)
>>
>> I know I can specify size of journal with -j option.
>>
>> Should the journal size be scaled in some proportion of size of UBIFS
>> partition?
> It does grow, but not infinitely. We have some heuristics in mkfs.ubifs,
> but those were mostly taken out of thin air. I do not remember us doing
> a lot of experimenting with that.
>
>> Should the journal size depend on how often you expect file-system to be
>> updated?
>>
>> Has anyone done experiments with setting really small (1MB/2MB) journal
>> size?
>> What side effects is one likely to encounter if the journal size is too
>> small.
>>
>> fwiw in my use case, file-system doesn't get updated that often, perhaps
>> 2K data in about 4 files once a week.
> I think you should do your own experiments. If you share the results -
> will be nice.
Yes, its on my to-do list once I am done resubmitting XATTR and integck 
patches.

-Subodh