On Thu, Oct 24, 2013 at 6:38 AM, Jukka Zitting <[email protected]> wrote:
> Hi,
>
> On Thu, Oct 24, 2013 at 2:48 AM, Tobias Bocanegra <[email protected]> wrote:
>> [Inline image 1]
>
> The image didn't survive it's trip to the mailing list. Can you post a
> link to it?
the relevant information was:

java.util.UUID, 353'574 Objects (41%), 22'628'736 shallow size

>
>> I thought that Oak does not use UUIDs so much anymore - why do I have 20mb 
>> of them? they are all allocated via the SegmentMK.
>
> The SegmentMK uses UUIDs to identify segments.
>
> Currently multiple copies of each UUID are kept in memory, one for the
> segment itself and another for each other segment that contains a
> reference to it. Avoiding those duplicates should bring the memory
> requirement down by an order of magnitude.
>
> Another reason for the high memory use is that currently the TarMK
> (not the MongoDB backend of SegmentMK) keeps all Segment UUIDs in
> memory all the time to avoid having to scan through the tar files when
> looking up a segment. A segment index at the end of each tar file
> should take care of that.
>
> BR,
>
> Jukka Zitting


--
toby

Reply via email to