Hi Emin,

I'm suggesting to use sorted vertex IDs when you insert the properties.

Thanks

Luigi



2015-06-19 17:20 GMT+02:00 Emin Agassi <[email protected]>:

> Hi Luigi,
>
> What do you mean by insert rate ? How long it takes to load this size of
> the graph?
> Currently, it takes 4 minutes to load 744,496 rows as Vertices
> and 6,445,621 rows as Edges.
> I am not sorting Vertix ids and not keeping the same order that I used for
> creating Edges.
> I would like to confirm that you are suggesting to use Vertix ids in the
> same order as used when creating Edges. Correct?
> Or, are you suggesting just sort the Vertix ids after I create them for
> the createEdges and use the sorted ids for setVertex properties?
>
> Also, my problem is that these newly generated Ids are not the ids used in
> the database. This means that I need to map between new generated ids and
> the object ids stored in the DB.
> So, I ended up having a HashMap between these new ids and ids coming from
> the DB. Otherwise, I do not know which object I am setting vertix
> properties for.
>
> Thanks
> Emin
>
>
> On Friday, June 19, 2015 at 10:25:18 AM UTC-4, Luigi Dell'Aquila wrote:
>>
>> Hi Emin,
>>
>> the batch insert is very fast because it does a lot of operations in RAM
>> and then flushes raw data to disk all together.
>> The maximum value of vertex IDs counts a lot, because OrientDB will
>> create (and in some cases destroy, during import process) as many records
>> as that number, so if you can keep it low you will have better performance.
>> The same is for the sorting order of vertex ids in set vertex properties,
>> I strongly suggest you to do fully sorted setVertexProperties().
>> Anyway, when you invoke setVertexProperties() records are actually
>> flushed to the clusters (before that everything happens in RAM), this is
>> why at that moment you have a slowdown.
>> Out of curiosity, which insert rate are you having?
>>
>> Thanks
>>
>> Luigi
>>
>>
>>
>>
>>
>> 2015-06-19 15:27 GMT+02:00 Emin Agassi <[email protected]>:
>>
>>>
>>> Hello Luigi or Luca,
>>>
>>> I have a question regarding the new Batch Insert class.
>>> I have a database that contains 744,496 rows which are loaded as
>>> Vertices and 6,445,621 rows which are loaded as Edges.
>>> I am using the new OGraphBatchInsert.
>>> I am following the procedure described in the Java comments: create
>>> edges first and then set Vertix properties.
>>> The createEdge() API requires two Long ids for Vertices.
>>> Today, I have an incremental counter that I use to generate these Ids.
>>> Then, I use the same ids in the set Vertix properties operation.
>>>
>>> Questions:
>>> Does it matter how large these generated Ids get?
>>> Should these ids be used in the same sequential order after creating
>>> them for the createEdge and then in the same order for the setVertix
>>> properties.
>>>
>>> Creating Edges is very fast for this size of the graph but setting
>>> Vertix properties is slow. Would you know why? Could this be related to the
>>> sorting order of the Vertex Ids between createEdge calls and set Vertix
>>> properties?
>>>
>>> I am not sure why set Vertix properties is so slower. I am not setting
>>> large properties. I only have 4 properties to set.
>>>
>>> Thank you for help
>>> Emin
>>>
>>>  --
>>>
>>> ---
>>> You received this message because you are subscribed to the Google
>>> Groups "OrientDB" group.
>>> To unsubscribe from this group and stop receiving emails from it, send
>>> an email to [email protected].
>>> For more options, visit https://groups.google.com/d/optout.
>>>
>>
>>  --
>
> ---
> You received this message because you are subscribed to the Google Groups
> "OrientDB" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"OrientDB" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to