1 |
My experience with NFS mounted roots is that they can bombard your |
2 |
network with packets. A simple little script can manage to generate |
3 |
enough traffic to actually slow down other services just by hitting |
4 |
tons of services. Plus if you have spoolers and such you end up |
5 |
generating a ton of traffic or memory. The advantage to a local |
6 |
install is that you can cut down on the network traffic drastically. |
7 |
Granted, if your applications are all embarrassingly parallel and |
8 |
don't do a ton of disk IO then NFS root works great.. Many of the |
9 |
applications we use here would utterly destroy the network if run |
10 |
from a NFS mounted root.. The advantage of rebuilding is consistency |
11 |
without the disadvantage of NFS roots. |
12 |
|
13 |
|
14 |
On Nov 21, 2005, at 9:45 AM, Eric Thibodeau wrote: |
15 |
|
16 |
> Le 21 Novembre 2005 11:10, Robin H. Johnson a écrit : |
17 |
>> On Sun, Nov 20, 2005 at 08:51:13PM -0500, St?phane Lacasse wrote: |
18 |
>> [snip discussion about installing] |
19 |
>> |
20 |
>> I've done the cluster system (128 node+ 1 master) in a similar |
21 |
>> fashion |
22 |
>> to what you are after. |
23 |
>> 1. PXE-boot install environment for performing installs of both the |
24 |
>> master and all of the nodes. |
25 |
> PXE-boot even for the Master?...so where do the images reside...how |
26 |
> do you |
27 |
> manage the slightly varying config items such as hostname and all? |
28 |
> This |
29 |
> approach still seems a little bit time consuming since all nodes |
30 |
> are still |
31 |
> individual entities (not NFS roots to a single maintained image). |
32 |
> Though |
33 |
> granted that the nodes being all identical, emerge -K should in |
34 |
> theory be a |
35 |
> breeze....but it's not the case for maintaining all the config files |
36 |
> consistent. |
37 |
> |
38 |
>> 2. The install environment uses the Gentoo Installer, with the CLI |
39 |
>> frontend I wrote for the GLI project, and performs complete |
40 |
>> installs of |
41 |
>> nodes in under 20 minutes (depending on network traffic). |
42 |
> So switching a machine's purpose/profile requires a complete re- |
43 |
> install on the |
44 |
> node? You state 20 minutes for re-installing, is it a _real_ |
45 |
> install or the |
46 |
> dump of a "reference" root? (Pardon my ignorance of the CLI |
47 |
> installer you are |
48 |
> referring to... I'll read the http link you'll send me ;) ) |
49 |
> |
50 |
>> By using GLI, it's a simple matter of altering the install |
51 |
>> profiles to |
52 |
>> reconfigure the cluster, and wipe the nodes for changing their |
53 |
>> purpose |
54 |
>> (presently we have an MPI mode and a MOSIX mode), some of the cluster |
55 |
>> users need assurances that none of their data remains on the cluster |
56 |
>> after they are done, hence being able to reinstall easily. |
57 |
> [...] |
58 |
>> Also, make use of your cluster tools to administer the cluster. |
59 |
>> OpenPBS |
60 |
>> allows running a job on all nodes, so use it to emerge -K [package]. |
61 |
>> (not -k as binpkgs don't currently have any locking in $PKGDIR, |
62 |
>> and can |
63 |
>> get corrupted if two emerge processes try to create a binpkg at the |
64 |
>> same time.) |
65 |
> |
66 |
> Actually, I would have thought you use _one_ node to compile the |
67 |
> packages |
68 |
> (using distcc at your description) and _then_ propagate the package |
69 |
> onto the |
70 |
> other nodes with -K....still, I would think maintaining an NFS |
71 |
> mounted ROOT |
72 |
> would be less cumbersome.... |
73 |
> |
74 |
> -- |
75 |
> Eric Thibodeau |
76 |
> |
77 |
> -- |
78 |
> gentoo-cluster@g.o mailing list |
79 |
> |
80 |
|
81 |
|
82 |
-- |
83 |
gentoo-cluster@g.o mailing list |