1 |
Michael Mol <mikemol <at> gmail.com> writes: |
2 |
|
3 |
|
4 |
> > http://shader.kaist.edu/packetshader/ |
5 |
> > anyone interested in trying? |
6 |
|
7 |
A firewall router based on a GPU+CPU is a great idea, who's |
8 |
stability is probably a few years away. |
9 |
|
10 |
Basis: GPU use very fast memory often with special features, |
11 |
based on architecture. But the proof is in the pudding; i.e. |
12 |
such a device being tested in a variety of scenarios. The basic |
13 |
problem is you have different details on GPU and often the |
14 |
necessary details of the hardware, are not published in a general |
15 |
access type of document. Most chip vendors, when they do get |
16 |
a software firewall router working on a chipset (GPU + CPU) |
17 |
will most likely want to sell a solution, rather than open |
18 |
source this solution. It has huge ramification commercially. |
19 |
|
20 |
So for this to be a fruitful effort, I'd suggest waiting until you |
21 |
have one of those fancy new AMD chips where the GPU and mutli-core |
22 |
CPU are on the same die PLUS and open source project. |
23 |
|
24 |
Intel has nice (CPU) hardware, but video is a pig (dog_slow) on |
25 |
the Intel GPU. The intel GPU is a DOG...... |
26 |
|
27 |
Nvidia has some nice software offerings, but no robust CPU multi |
28 |
core to work with the GPU (on the same die). Also, Nvidia has a |
29 |
weak history of open-source support. In fact the project you |
30 |
mention may not even publish other parts of the sourcecode, according |
31 |
to the website. So why would you waste your time on that code offering. |
32 |
|
33 |
|
34 |
When somebody (GPU team) get's iptables and gentoo running on a new, |
35 |
integrated (GPU + CPU) AMD chip, just use vanilla tools via |
36 |
the gentoo organization? Then IP tables just has to be modified |
37 |
to take advantage of the GPU. Maybe GCC will handle this some day. |
38 |
maybe AMD will open source some internal knowledge to make it |
39 |
happen; Maybe not. |
40 |
|
41 |
|
42 |
> I see a lot of graphs touting high throughput, but what about latency? |
43 |
|
44 |
This is a good point. Wait until the GPU and CPU are on the same |
45 |
die... (think AMD). I just do not see Intel or Nvidia being the first |
46 |
to make this truely a commodity (too much money to made selling the |
47 |
proprietary solutions). For example, and high-end compiler vendor |
48 |
could buy an exclusive license from Intel/Nvidia, to make this |
49 |
a unique and expensive offering. Think DOD contractors that |
50 |
limit the solution to VME buss based systems (just a random thought). |
51 |
|
52 |
> They also tout a huge preallocated packet buffer, and I'm not sure |
53 |
> that's a good thing, either. It may or may not cause latency problems, |
54 |
> depending on how they use it. |
55 |
|
56 |
Traditionally, searching and sorting algos smoke on GPU and GPU |
57 |
type ram. Other processes not so wonderful. That's why you need |
58 |
the GPU to offload processes, that it can run much faster than |
59 |
the multi-core CPU. Both are needed most of the time. |
60 |
|
61 |
> They don't talk about latency at all, except for one sentence: |
62 |
|
63 |
I think that site is just trying to get folks to do some testing |
64 |
for them. They do not seem to be 'open-source' minded, imho.... |
65 |
|
66 |
Personally, I would not waist my time. But do watch out for new |
67 |
offerings from AMD..... Maybe Intel (naw, just kidding, they |
68 |
never release or support anything open source, until they |
69 |
have to....) I bet those developers had to sign some serious |
70 |
NDAs with some nasty corporate types of lawyers.... |
71 |
|
72 |
just my opinion |
73 |
hth, |
74 |
James |