1 |
On Sat, 17 Jan 2015 21:04:44 -0500 Rich Freeman wrote: |
2 |
> Speak for yourself. :) I did comment on my thoughts in this area in |
3 |
> Donnie's thread. Gentoo (IMHO) tends not to be the best distro for |
4 |
> doing anything in particular. I find that its best feature is that it |
5 |
> is reasonably good at doing just about anything - it is a |
6 |
> jack-of-all-trades. |
7 |
|
8 |
I can't agree with you here, though your position have a rationale. |
9 |
I see Gentoo as a Universal Constructor (UC) which may be used to |
10 |
whatever specific needs Linux can be used at all. |
11 |
|
12 |
In general UC pros is ability to create setup suitable for every |
13 |
specific need, but cons is maintenance cost to create and update |
14 |
such setup. Also creating and maintaining UC-powered setups rises |
15 |
general professional level of system architect or amdin doing the |
16 |
job. |
17 |
|
18 |
So everything comes to how much user needs deviate from what |
19 |
already existing binary distributions provide. If user needs are |
20 |
perfectly satisfied with some binary distro, using Gentoo will only |
21 |
raise maintenance costs. But if users demands something hardly |
22 |
achievable with other (binary) distributions, then this is a good |
23 |
place for Gentoo. |
24 |
|
25 |
From my own experience I can point three directions where Gentoo |
26 |
was and is reasonably the best choise for our needs (mine or my |
27 |
colleagues): |
28 |
|
29 |
1) HPC. When it comes to scalable tasks and large amount of |
30 |
hardware, even small performance gain results into huge saving of |
31 |
costs. On our first cluster we replaced CentOS by carefully |
32 |
tuned Gentoo and performance gain was about 30-50% depending on |
33 |
scientific application (please note I'm talking about real |
34 |
applications and not about synthetic tests like linpack). With |
35 |
hardware costs about million of dollars, 30% performance gain |
36 |
results in a great saving. Price for that was much longer time for |
37 |
initial setup (many weeks instead of many days), but it was |
38 |
still less then time required to setup hardware itself and all |
39 |
auxiliary engineering systems. |
40 |
|
41 |
An interesting observation here is that average software update |
42 |
cost of Gentoo is smaller that one of RH-based systems we used |
43 |
before. While it is easier to update RH-based solution within the |
44 |
same branch, then Gentoo setup, it is a complete nightmare to |
45 |
upgrade from one branch to another, e.g. from RHEL4 to RHEL5. I've |
46 |
gone through such update in the past an it is much worse than remove |
47 |
everything and install from scratch, including all user |
48 |
applications. As for Gentoo, all updates are equal: they bring some |
49 |
build failures, runtime issues and compatibility problems, but to |
50 |
a limited extent, which is handleable easy enough by prepared team. |
51 |
|
52 |
2) High security servers. We have some systems dedicated to a very |
53 |
specific needs where security demands are extreme. Hardened Gentoo |
54 |
is the best solution here, since we can strip down such system close |
55 |
to an absolutely possible minimum and protect that minimum by all |
56 |
means (hardened toolchain and flags, PaX, SELinux and so on). Of |
57 |
course, on top of then containers may be use to isolate different |
58 |
daemons and so on... |
59 |
|
60 |
3) Individual interested in getting every bit of performance |
61 |
possible from own hardware. Frankly this was the reason why I |
62 |
switched to Gentoo from RH about 8 years ago. I just tired to |
63 |
rebuild each time a significant part of packages with custom flags |
64 |
and configure options. Gentoo is much better suited for this task. |
65 |
And as a result 13 years old hardware is still usable to watch 720p |
66 |
and most of 1080p videos (without GPU hardware decoding). A |
67 |
byproduct of such interest is a deep understanding of system |
68 |
internals, which is a great result on its own. |
69 |
|
70 |
Best regards, |
71 |
Andrew Savchenko |