1 |
On Thu, Dec 22, 2011 at 10:55 AM, Michał Górny <mgorny@g.o> wrote: |
2 |
>> Just wanted to point out that (if there is enough memory) recent |
3 |
>> kernels manage much better parallelism, even excess of it, once |
4 |
>> reached the maximum load augmenting threads only bring minimal loss of |
5 |
>> "real" time. |
6 |
> |
7 |
> Does that include handling complete lack of memory and heavy swapping? |
8 |
> |
9 |
|
10 |
I think the key was the "if there is enough memory" - which I think is |
11 |
a pretty big issue. Running 25 invocations of java each wanting |
12 |
100-200MB of RAM will cripple a system relatively quickly. And 25 |
13 |
really is just scratching the surface when most people think |
14 |
"cluster." |
15 |
|
16 |
I think the only way to really fix this would be to integrate |
17 |
distributed behavior into make itself. |
18 |
|
19 |
Rich |