| 1 |
On Fri, Sep 23, 2011 at 12:06 AM, Pandu Poluan <pandu@××××××.info> wrote: |
| 2 |
> Saw this on the pfSense list: |
| 3 |
> |
| 4 |
> http://shader.kaist.edu/packetshader/ |
| 5 |
> |
| 6 |
> anyone interested in trying? |
| 7 |
|
| 8 |
I see a lot of graphs touting high throughput, but what about latency? |
| 9 |
That's the kind of stuff that gets in my way when I'm messing with |
| 10 |
things like VOIP. |
| 11 |
|
| 12 |
My first thought when I saw they were using a GPU for processing was |
| 13 |
concerns about latency: |
| 14 |
1) RTT between a video card and the CPU will cause an increase in |
| 15 |
latency from doing processing on-CPU. Maybe DMA between the video card |
| 16 |
and NICs could help with this, but I don't know. Certainly newer CPUs |
| 17 |
with on-die GPUs will have an advantage here. |
| 18 |
2) GPGPU coding favors batch processing over small streams. That's |
| 19 |
part of its nature, after all. That means that processed packets would |
| 20 |
come out of the GPU side of the engine in bursts. |
| 21 |
|
| 22 |
They also tout a huge preallocated packet buffer, and I'm not sure |
| 23 |
that's a good thing, either. It may or may not cause latency problems, |
| 24 |
depending on how they use it. |
| 25 |
|
| 26 |
They don't talk about latency at all, except for one sentence: |
| 27 |
"Forwarding table lookup is highly memory-intensive, and GPU can |
| 28 |
acclerate it with both latency hiding capability and bandwidth." |
| 29 |
|
| 30 |
-- |
| 31 |
:wq |
Archives