1 |
On Tue, Aug 12, 2014 at 11:40 PM, Marc Joliet <marcec@×××.de> wrote: |
2 |
> Hi list |
3 |
> |
4 |
> For the longest time I've had portage configured to use the sqlite metadata |
5 |
> cache backend as per an old HOWTO [0], however, I thought that it would be a |
6 |
> good idea to revisit that decision. |
7 |
> |
8 |
> Now apparently, this was supposed to speed up portage, although even that |
9 |
> depends. For instance, [0] says that the metadata_overlay backend is faster on |
10 |
> fast storage; since all of portage is on an SSD, that ought to be the case for |
11 |
> me. However, [0] is pretty outdated, so I don't really know, and don't have any |
12 |
> comparison. |
13 |
> |
14 |
> In addition to that, I don't explicitly make use of the sqlite metadata cache, |
15 |
> that is, I don't (consciously) use any software that accesses those DBs, except |
16 |
> for eix (except for overlays, where one would need to run "emerge --regen" |
17 |
> first, which is *ssssslllllloooowwwww*), which can make use of them if |
18 |
> CACHE_METHOD is set appropriately; this speeds up eix-update considerably. |
19 |
> |
20 |
> Does anybody here have experience with this, or a recommendation? I tried |
21 |
> switching to the default cache method temporarily to see how things perform, |
22 |
> and "emerge @world -uDNva" slowed down by about 30 seconds, so preliminary |
23 |
> results point to sticking with sqlite (although it could at least partly be a |
24 |
> btrfs performance regression in Linux 3.15, since there have been several reports of |
25 |
> those, and several fixes slated for 3.16). Anyway, I'm also unsure of unintended |
26 |
> consequences, or other settings I might have to change, too. |
27 |
> |
28 |
> Also, does anybody have any performance data and/or experience on using btrfs |
29 |
> with compression in this context? |
30 |
> |
31 |
> [0] http://www.gentoo-wiki.info/TIP_speed_up_portage_with_sqlite |
32 |
> |
33 |
> Greetings, |
34 |
> -- |
35 |
> Marc Joliet |
36 |
> -- |
37 |
> "People who think they know everything really annoy those of us who know we |
38 |
> don't" - Bjarne Stroustrup |
39 |
|
40 |
Having tried this feature, I'd advise against it. It takes long time |
41 |
to generate metadata after sync and not really that advantageous. Also |
42 |
eix has it's own issues in this mode. |