That answers that question. We were running in .5G.

On Wed, Jun 13, 2012 at 7:52 AM, Andy Seaborne <[email protected]> wrote:
> On 13/06/12 12:13, Benson Margulies wrote:
>>
>> On Wed, Jun 13, 2012 at 5:20 AM, Damian Steer<[email protected]>
>>  wrote:
>>>
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> On 12/06/12 18:47, Benson Margulies wrote:
>>>>
>>>> I hit the problem I sent mail about most recently whilst trying to
>>>> reproduce the following. Is there any reason to think that this is
>>>> worse than TDB being unlucky enough to be the allocation straw
>>>> that breaks the camel's back?
>>>
>>>
>>> What are you running this on? On 64bit machines TDB should be
>>> reasonably light on the heap.
>>
>>
>> "A whole lot of triples"? It's sitting there stuffing triples into the
>> store, some, as you see, are reifications, which results in some
>> queries, and in other cases we're explicitly querying to decide what
>> to insert. Running the whole business with -Xmx1g succeeds. If there
>> is some more specific characterization of the process that would help,
>> please let me know. I obviously can't deliver the entire circus as a
>> test case.
>
>
> You are running 64bit presumably.  1G should be enough for TDB 9bit on the
> small side?) but of course it's competing for space with the rest of the
> app.
>
> On the surface, TDB is the unlucky straw, although realistically it is using
> a decent proportion of that 1G so it's putting itself in the firing line.
>
> The main heap memory consumer on a 64bit machine (mapped mode) is the node
> table.
>
> I normally run with 1.5G.
>
>        Andy
>

Reply via email to