| 1 |
Hi Eric, |
| 2 |
|
| 3 |
Sorry, forgot to CC the list on this. Here goes... |
| 4 |
|
| 5 |
On 4/11/06, Eric Thibodeau <kyron@××××××××.com> wrote: |
| 6 |
> Mr. Rand(), |
| 7 |
> |
| 8 |
> I tend to prone the Diskless approach for the reasons you are mentionning. |
| 9 |
> You can easily switch between roots with a network boot just by modifying |
| 10 |
> your dhcp config and rebooting a node. This is really neat since you can have |
| 11 |
> the dev environment on the actual cluster, test the new root with new libs |
| 12 |
> and simply reboot some available nodes to test them. Furthermore, this |
| 13 |
> approach opens the way to having multiple boot profiles with |
| 14 |
> application-specific orientations (/me is thinking of the hellish deal of |
| 15 |
> parallel Matlab with a polluted environment and the booting into a really |
| 16 |
> optimized one for real MPI work ;)...) |
| 17 |
|
| 18 |
Yes, this is exactly what I had in mind :) |
| 19 |
|
| 20 |
> As for the "un-ncecessary attacker blahblahblah... Put your head behing the |
| 21 |
> firewall. Beowulf nodes aren't meant to be publically available if they are |
| 22 |
> to be efficient. nonetheless, departmental clusters (by night) could aslo be |
| 23 |
> very possible with the diskless approach |
| 24 |
|
| 25 |
Sorry, I should have mentioned that my application is load balancing / |
| 26 |
HA rather than HPC. There will certainly be some very strict firewall |
| 27 |
rulesets in place, however I do have to deal with serving up user |
| 28 |
requests. If someone manages to poke a hole in Apache or whatever and |
| 29 |
gets themself a shell I don't want to give them any tools which might |
| 30 |
help them along their quest for root. |
| 31 |
|
| 32 |
-- |
| 33 |
gentoo-cluster@g.o mailing list |
Archives