1 |
On Thursday 10 April 2003 02:44 am, Eske Christiansen wrote: |
2 |
> On Thu, Apr 10, 2003 at 07:06:28AM +0800, William Kenworthy wrote: |
3 |
> > But is it a "real" problem or just the way you have configured the |
4 |
> > kernel? One respondent suggested low latency: have you configured the |
5 |
> > kernel the same as the redhat one? Is large memory selected in gentoo |
6 |
> > and not redhat etc? |
7 |
> |
8 |
> Don't you think that a syscall to tcp select that is 5 times slower is a |
9 |
> problem. I do. |
10 |
|
11 |
Not always, if reliability goes 5x it's well worth it to me in some cases. |
12 |
|
13 |
> |
14 |
> > Are you going to use it as a server or workstation: I would look at |
15 |
> > first benchmarking same against same, then configuring gentoo for the |
16 |
> > task and look at any advantages (e.g., low latency for instance, not |
17 |
> > just raw speed which is more a server thing) you get there. Does redhat |
18 |
> > do Low Latency as standard in its workstation kernels? |
19 |
> |
20 |
> As I stated it, it is used in a cluster where the conection is a giga- |
21 |
> bit ethernet. In that case I want TCP select to as fast as it can be. |
22 |
> I can see that redhat can make i 5 times faster. |
23 |
> If the gentoo kernel is not best performance kernel that is offered in |
24 |
> gentoo don't you think that an update to the installation text is in order. |
25 |
|
26 |
Sometimes speed comes at a sacrifice to other things notably stability and |
27 |
cleanliness(see recent discussions of vendor kernels on LKML archives to see |
28 |
Red Hat emplyees bad mouthing the cleanliness of RH kernels). Not everybody |
29 |
is solely interested in speed. I have made compromises on most of my servers |
30 |
that lean them more toward stability, because I don't like getting pages in |
31 |
the middle of the night because some kernel patch wasn't exactly right. It |
32 |
all depends on what you need. That is why most people are drawn to Gentoo. |
33 |
The installation documanetation should be taken as a bare minimum to get a |
34 |
system up and running, from there you are on your own to find out what suits |
35 |
you best. I personally wouldn't use the RH sources if somebody paid me, they |
36 |
change too many things. They can't keep up with all that stuff sometimes(no |
37 |
offense to the RH guys, I know it is a full time job just reading LKML not to |
38 |
mention getting some other work done during the day) |
39 |
|
40 |
> |
41 |
> What we could do in gentoo is to make a gentoo-soruces-workstation and a |
42 |
> gentoo-sources-server, so we can get a kernel that fit the workstation use |
43 |
> and a kernel optimized for raw speed. |
44 |
|
45 |
There are already tons of different patchsets out there, people would be |
46 |
better off using one of them. I use aa kernels on most of my servers and a |
47 |
mixture of ck and wolk sources on most of my desktops. |
48 |
|
49 |
> |
50 |
> eske |
51 |
<snip> |
52 |
|
53 |
-- |
54 |
OpenGFS -- http://opengfs.sourceforge.net |
55 |
Home -- http://www.brianandsara.net/brian |
56 |
|
57 |
-- |
58 |
gentoo-dev@g.o mailing list |