1 |
Tom H <tomh0665 <at> gmail.com> writes: |
2 |
|
3 |
|
4 |
> >> /usr/src/linux/scripts/kconfig/Makefile |
5 |
> >> /usr/src/linux/scripts/kconfig/conf.c |
6 |
|
7 |
> I'm not too sure what you're looking for but if you want to compare |
8 |
> config_new and config_old where you generated config_new with |
9 |
> config_old and "make oldconfig", you can use "./scripts/diffconfig |
10 |
> .config config.old". |
11 |
|
12 |
|
13 |
Background:: |
14 |
I have been researching and reading about a plethora of cluster/cloud |
15 |
approaches to running normal linux codes (frameworks if you like) and also |
16 |
Hi Performance Computing (many machines on a single problem). Long story |
17 |
short version, it seems the more the linux system is minimized, in size and |
18 |
complexity, the faster the containers or HPC applications run on top of it |
19 |
and the easier it is to 'secure' the cluster. Many, in the cluster world, |
20 |
particularly the HPC folks, are not using systemd as it adds unnecessary |
21 |
complexity and degrades performance, albeit systemd does bring some ease of |
22 |
management to large numbers of nodes. Gentoo has quite a following of openrc |
23 |
folks, as I am one, and there needs to be a openrc pathway to |
24 |
cluser-paradise, imho. Most folks are keeping their secrets as to how to |
25 |
best tune a linux kernel for cluster or container deployment. Mix that in |
26 |
with Systemd complexities and the result is a very obfuscated environment |
27 |
for kernel tuning and benchmarking of clusters. So I'm going to need tools |
28 |
to rapidly (CI?) rebuild various linux kernels with minor tweaks to the |
29 |
config, as well as deployment consistency. |
30 |
|
31 |
Cluster vendors have basically two strategies:: Commercially offer a version |
32 |
of the cluster that is tuned to a specific performance need, and/or |
33 |
commercially offer the very nice management tools that make clustering |
34 |
easy, pleasant and robust. |
35 |
|
36 |
|
37 |
I have become very interested in developing a small cluster for testing some |
38 |
of the myriad of cluster offerings. IMHO:: a cluster is just a local cloud |
39 |
that is run locally. So for my example 12 systems, mostly amd64, but some |
40 |
intel and arm64 systems, are to be used. The desire is to be able to use |
41 |
these systems, to rapidly throw up a cluster for containers, or HPC, run |
42 |
some codes, trap and retain data for later analysis and comparison with runs |
43 |
against other cluster architectures and codes that constitute a given target |
44 |
cluster. Loosely, cluster benchmarking. |
45 |
|
46 |
|
47 |
There is the myriad of 'frameworks' that are available for the various |
48 |
clusters one can build. It's all quite complex on the surface (vendor |
49 |
hyperbole), but underneath, vetting performance claims is quite simple. Load |
50 |
of the test-cluster with codes, run some jobs/apps/codes on a given |
51 |
cluser+frameworks, collect data for analysis and comparison. |
52 |
|
53 |
|
54 |
Present:: |
55 |
Gentoo's GSoC has at least two projects that are very appealing to me along |
56 |
these lines:: |
57 |
|
58 |
1) Stateless Minimal Gentoo |
59 |
|
60 |
2) kernelconfig |
61 |
|
62 |
Item two is the tie-in for oldconfig and associated semantics. PXE boot and |
63 |
other metrics are at play here too. So a comprehensive kernel build and |
64 |
deploy system, for lots of systems and different architectures, is of |
65 |
interest. Most prototype work is done on amd64. Live kernel patching is |
66 |
an active area of development and in some circumstances will not necessitate |
67 |
the reboot of the system/node to update the kernel [A]. |
68 |
|
69 |
Item one is similar to CoreOS. CoreOS has a system for rapidly deploying |
70 |
systems from a cold boot as well as live updates to the running OS with |
71 |
rollback if necessary. Impressive to say the least. But, CoreOS is locked |
72 |
into the systemd pathway. Myself, being of the minimalistic embedded ilk, I |
73 |
tend to lean towards the alpine-docker pathway of minimization and |
74 |
openrc/busybox. |
75 |
|
76 |
|
77 |
Lofty goals. But, extraordinarily useful to linux folks, especially those |
78 |
with small to medium size data-centers. Note, lots of folks drop me email |
79 |
privately with information, concerns and tidbits of wisdom most useful |
80 |
in my journey to cluster-paradise. Public guidance and even scorn, are also |
81 |
most welcome. Reams of inexpensive arm64 machines combined with compiler |
82 |
advances make clustering the most exciting adventure in the linux world, atm. |
83 |
|
84 |
|
85 |
hth, |
86 |
James |
87 |
|
88 |
[A] http://rhelblog.redhat.com/2015/03/23/live-kernel-patching-update/ |