Dropping mode optimization from Metafile?
robb at canfield.com
Wed Aug 29 01:38:29 UTC 2007
I currently hacked in the "standard" permissions for files and dirs
where standard is what my 3 distributions use! Other solutions are
possible but I can't help but wonder if its 90% work for for 1% gain, so
I am not pursuing them right now.
Richard Edward Horner wrote:
> I'd say bring this up again when Brad gets back from camping or
> whatever because he probably has reasons for this architecture but I
> do agree that the issues you point out are unfortunate.
> Thanks, Rich(ard)
> On 8/28/07, robb <robb at canfield.com> wrote:
>> During a test of a medium scale backup (simulating RAM and directory
>> overhead for 77,000 files averaging 1 MB) I am finding some higher than
>> desired RAM usage.
>> Brackup buffers all of the file stats in RAM (something that is a bit
>> tricky to change). This is a constant for the run, but Brackup also
>> buffers all chunk data until the backup is done and the metafile
>> written. This chunk data can add up to a considerable amount of usage.
>> The main reason for buffering chunk data seems to be to optimize the
>> size of the file by remembering the most common file and directory
>> modes. If the file/dir is the default mode then the mode line need not
>> be added.
>> While optimizing the size of a file is always nice, the downsides for
>> large backups may overshadow this:
>> - A failure during backup leaves then brackup metafile out of date (or
>> non-existent). Even if MANY files are properly saved,, the index to them
>> is not available.
>> - RAM is consumed and CPU resources used for saving this data in RAM.
>> This becomes an issue on virtualized systems with low RAM thresholds OR
>> for any system with LARGE data sets. Brackup currently stores the entire
>> file list in RAM, but during backup the chunk references are added to
>> this and it is those chunk references that start to add up for large
>> files or large backup sets and small a small chunk size makes it even worse.
>> - The optimization is relatively small (actually very small) size wise
>> given the other data in the file.
>> I vote to drop the optimization file size and favor RAM (and execution
>> speed) by streaming the metafile as each file is finished. Any thoughts?
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 3237 bytes
Desc: S/MIME Cryptographic Signature
Url : http://lists.danga.com/pipermail/brackup/attachments/20070828/24ef04ad/smime.bin
More information about the brackup