Home » Eclipse Projects » GEF » GEF Benchmarks
| GEF Benchmarks [message #82510] |
Thu, 05 June 2003 18:38  |
Eclipse User |
|
|
|
Hello,
I am considering using GEF/draw2d for a Graphing Library. In order to
evaluate its suitability I would like to know what the biggest graph I
could make is.
I am a biologist, so I would need to be able to manipulate on the order
of 100,000 nodes ~1/4 connected. Ideally I would be able to zoom in and
manipulate individual nodes, and then zoom out for a whole-network view.
If anybody has any ideas on this sort of thing, please let me know.
--rowan
|
|
|
| Re: GEF Benchmarks [message #82540 is a reply to message #82510] |
Thu, 05 June 2003 23:10  |
| Eclipse User |
|
|
|
Originally posted by: none.us.ibm.com
The implementation of Figure has fields for children, borders, layouts,
minimum sizes, etc. So the memory footprint of a figure could be smaller if
it was designed not to be a container. I just tried instantiating 50,000
RectangleFigures, and the memory requirement was 42MB. With zoom, the
performance starts to get pretty bad, I think because garbage collection was
being forced during the painting.
What is the level of interaction that you need with this graph? Do you need
to allow the user to edit anything, or is this read-only data? If it is
just display data, you should write a single figure implementation that
paints itself based on a model. That model would be a UI model containing
the 100,000 nodes. This way, you could still leverage the zoom capability
in draw2d.
Also, you might also implement a BSP tree for better painting performance.
"Rowan Christmas" <xmas@systemsbiology.org> wrote in message
news:bbogpd$tjb$1@rogue.oti.com...
> Hello,
>
> I am considering using GEF/draw2d for a Graphing Library. In order to
> evaluate its suitability I would like to know what the biggest graph I
> could make is.
>
> I am a biologist, so I would need to be able to manipulate on the order
> of 100,000 nodes ~1/4 connected. Ideally I would be able to zoom in and
> manipulate individual nodes, and then zoom out for a whole-network view.
>
> If anybody has any ideas on this sort of thing, please let me know.
>
> --rowan
>
|
|
|
| Re: GEF Benchmarks [message #83027 is a reply to message #82540] |
Tue, 10 June 2003 14:21 |
| Eclipse User |
|
|
|
Thanks for replying, I was away the past few days.
I would like to be able to have a very interactive display, like Piccolo
in the Swing world. Although it becoming apparent to me that I can only
reasonably support interaction for >5,000 nodes. At higher numbers the
one figure approach is what I will do.
I will look into BSP trees, but being primarily a biochemist I have only
limited familiarity with some of these complex data structures.
How were you thinking that a BSP tree would be used for painting? It
seems that I would just use a BFS from my graph data structure to draw
the graph.
--rowan
Randy Hudson wrote:
> The implementation of Figure has fields for children, borders, layouts,
> minimum sizes, etc. So the memory footprint of a figure could be smaller if
> it was designed not to be a container. I just tried instantiating 50,000
> RectangleFigures, and the memory requirement was 42MB. With zoom, the
> performance starts to get pretty bad, I think because garbage collection was
> being forced during the painting.
>
> What is the level of interaction that you need with this graph? Do you need
> to allow the user to edit anything, or is this read-only data? If it is
> just display data, you should write a single figure implementation that
> paints itself based on a model. That model would be a UI model containing
> the 100,000 nodes. This way, you could still leverage the zoom capability
> in draw2d.
>
> Also, you might also implement a BSP tree for better painting performance.
>
> "Rowan Christmas" <xmas@systemsbiology.org> wrote in message
> news:bbogpd$tjb$1@rogue.oti.com...
>
>>Hello,
>>
>>I am considering using GEF/draw2d for a Graphing Library. In order to
>>evaluate its suitability I would like to know what the biggest graph I
>>could make is.
>>
>>I am a biologist, so I would need to be able to manipulate on the order
>>of 100,000 nodes ~1/4 connected. Ideally I would be able to zoom in and
>>manipulate individual nodes, and then zoom out for a whole-network view.
>>
>>If anybody has any ideas on this sort of thing, please let me know.
>>
>>--rowan
>>
>
>
>
|
|
|
| Re: GEF Benchmarks [message #83113 is a reply to message #83027] |
Tue, 10 June 2003 15:17 |
| Eclipse User |
|
|
|
Originally posted by: none.us.ibm.com
I was just wondering if node.intersects(updateRegion) would get expensive
for 100,000 nodes.
If it does, you can partition the nodes into visibility regions. Then, if
the entire region is not visible, you can ignore a large number of nodes.
It's best to just write it first and find out the performance behavior/.
"Rowan Christmas" <xmas@systemsbiology.org> wrote in message
news:bc57jg$36s$1@rogue.oti.com...
> Thanks for replying, I was away the past few days.
>
> I would like to be able to have a very interactive display, like Piccolo
> in the Swing world. Although it becoming apparent to me that I can only
> reasonably support interaction for >5,000 nodes. At higher numbers the
> one figure approach is what I will do.
>
> I will look into BSP trees, but being primarily a biochemist I have only
> limited familiarity with some of these complex data structures.
>
> How were you thinking that a BSP tree would be used for painting? It
> seems that I would just use a BFS from my graph data structure to draw
> the graph.
>
> --rowan
>
>
> Randy Hudson wrote:
> > The implementation of Figure has fields for children, borders, layouts,
> > minimum sizes, etc. So the memory footprint of a figure could be
smaller if
> > it was designed not to be a container. I just tried instantiating
50,000
> > RectangleFigures, and the memory requirement was 42MB. With zoom, the
> > performance starts to get pretty bad, I think because garbage collection
was
> > being forced during the painting.
> >
> > What is the level of interaction that you need with this graph? Do you
need
> > to allow the user to edit anything, or is this read-only data? If it is
> > just display data, you should write a single figure implementation that
> > paints itself based on a model. That model would be a UI model
containing
> > the 100,000 nodes. This way, you could still leverage the zoom
capability
> > in draw2d.
> >
> > Also, you might also implement a BSP tree for better painting
performance.
> >
> > "Rowan Christmas" <xmas@systemsbiology.org> wrote in message
> > news:bbogpd$tjb$1@rogue.oti.com...
> >
> >>Hello,
> >>
> >>I am considering using GEF/draw2d for a Graphing Library. In order to
> >>evaluate its suitability I would like to know what the biggest graph I
> >>could make is.
> >>
> >>I am a biologist, so I would need to be able to manipulate on the order
> >>of 100,000 nodes ~1/4 connected. Ideally I would be able to zoom in and
> >>manipulate individual nodes, and then zoom out for a whole-network view.
> >>
> >>If anybody has any ideas on this sort of thing, please let me know.
> >>
> >>--rowan
> >>
> >
> >
> >
>
|
|
|
Goto Forum:
Current Time: Fri Nov 07 08:00:32 EST 2025
Powered by FUDForum. Page generated in 0.05768 seconds |
Home