Re: Overlap the execution of SPU?

Daniel wrote:
> Hello,
> 
> I'm using Chromium in a 10 nodes cluster with one client and nine
> servers which form a 3x3 passive stereo display. Each node is
> equipt with double Xeon 2.4G CPU, 1G memory, GeForce 5200 with
> PCI-X bus. They are connected by Giga Ethernet.
> 
> However, the speed is always below my expectation. Rendering of a
> moderate size object with about 60K triangles needs about 0.6~1
> seconds, which only need 0.04~0.05s on a single nodes.

You're probably limited by the network.  One potential work-around is 
to use display lists.  It's best to only put geometry and not 
state-change commands in the display list.


> I analyzed the problem carefully. I do think there are several
> issues may be considered to optimize Chromium. One possible way is
> overlaping the execution of SPU. If the network send/rec and
> rendering can be overlapped, then lots of time will be saved. And
> if every SPU spawns one thread, the stream may be parallized to
> utilize the SMP nodes, may it?

That's something we'd like to do, but it would involve some work in 
the network layer.


> Another problem: I find that Chromium doesn't consider load balance
> in Tiled display wall. Can we add one in it?

There's a Chromium extension (GL_CR_tile_info) that lets you change 
the tile sizes at runtime.  Chromium doesn't have any built-in load 
balancing, but you could implement it in your app.

-Brian

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV