1 |
Am Wed, 24 May 2017 08:00:33 +0200 |
2 |
schrieb Kai Krakow <hurikhan77@×××××.com>: |
3 |
|
4 |
> Am Wed, 24 May 2017 07:34:34 +0200 |
5 |
> schrieb gentoo-user@××××.de: |
6 |
> |
7 |
> > On 17-05-23 at 22:16, Ian Zimmerman wrote: |
8 |
> > > So what are gentoo users' opinions on this matter of faith? |
9 |
> > I use an ext4 partition backed by zram. Gives me ~3x compression on |
10 |
> > the things I normally have lying around there (plain text files) and |
11 |
> > ensures that anything I throw there (or programs throw there) gets |
12 |
> > cleaned up on reboot. |
13 |
> > |
14 |
> > > I have long been in the camp that thinks tmpfs for /tmp has no |
15 |
> > > advantages (and may have disadvantages) over a normal filesystem |
16 |
> > > like ext3, because the files there are normally so small that they |
17 |
> > > will stay in the page cache 100% of the time. |
18 |
> > I've never actually benchmarked this. Most of the things I notice |
19 |
> > that tend to end up there are temporary build files generated during |
20 |
> > configure stages or temporary log files used by various programs |
21 |
> > (clang static analyzer). Even if the entire file stays in the page |
22 |
> > cache, it'll still generate IO overhead and extra seeks that might |
23 |
> > slow down the rest of your system (unless your /tmp is on a |
24 |
> > different hard drive) which on spinning rust will cause slowdowns |
25 |
> > while on an ssd it'll eat away at your writes (which you may or may |
26 |
> > not have to worry about). |
27 |
> > |
28 |
> > > But I see that tmpfs is the default with systemd. Surely they |
29 |
> > > have a good reason for this? :) |
30 |
> > Or someone decided they liked the idea and made it the default and |
31 |
> > nobody ever complained (or if they did were told to just change it |
32 |
> > on their system). |
33 |
> > |
34 |
> > Either way, it'd be nice if someone actually benchmarked this. |
35 |
> |
36 |
> While I have no benchmarks and use the systemd default of tmpfs |
37 |
> for /tmp, I also put /var/tmp/portage on tmpfs, automounted through |
38 |
> systemd so it is cleaned up when no longer used (by unmounting). |
39 |
> |
40 |
> What can I say? It works so much faster: Building packages is a lot |
41 |
> faster most of the time, even if you'd expect gcc uses a lot of |
42 |
> memory. |
43 |
> |
44 |
> Well, why might that be? First, tmpfs is backed by swap space, that |
45 |
> means, you need a swap partition of course. |
46 |
|
47 |
To get in line with Rich Freeman: I didn't want to imply that zswap |
48 |
only works with swap, neither that tmpfs only works with swap. Both |
49 |
work without. But if you want to put some serious amount of data into |
50 |
tmpfs, you need swap as a backing device sooner or later. |
51 |
|
52 |
> Swap is a lot simpler than |
53 |
> file systems, so swapping out unused temporary files is fast and is a |
54 |
> good thing. Also, unused memory sitting around may be swapped out |
55 |
> early. Why would you want inactive memory resident? So this is also a |
56 |
> good thing. Portage can use memory much more efficient by this. |
57 |
> |
58 |
> Applying this reasoning over to /tmp should no explain why it works so |
59 |
> well and why you may want it. |
60 |
> |
61 |
> BTW: I also use zswap, so tmpfs sits in front of a compressed |
62 |
> write-back cache before being written out to swap compressed. This |
63 |
> should generally be much more efficient (performance-wise) than |
64 |
> putting /tmp on zram. |
65 |
> |
66 |
> I configured tmpfs for portage to use up to 30GB of space, which is |
67 |
> almost twice the RAM I have. And it works because tmpfs is not |
68 |
> required to be resident all the time: Inactive parts will be swapped |
69 |
> out. The kernel handles this much similar to the page cache, with the |
70 |
> difference that your files aren't backed by your normal file system |
71 |
> but by swap. And swap has a lot lower IO overhead. |
72 |
> |
73 |
> Overall, having less IO overhead (and less head movement for portage |
74 |
> builds) is a very very efficient thing to do. GCC constantly needs all |
75 |
> sorts of files from your installation (libs for linking, header files, |
76 |
> etc), and writes a lot of transient files which are needed once later |
77 |
> and then discarded. There's no point in putting it on a non-transient |
78 |
> file system. |
79 |
> |
80 |
> I use the following measures to get more performance out of this |
81 |
> setup: |
82 |
> |
83 |
> * I have three swap partitions spread across three HDDs |
84 |
> * I have a lot of swap space (60 GB) to have space for tmpfs |
85 |
> * I have bcache in front of my HDD filesystem |
86 |
> * I have a relatively big SSD dedicated to bcache |
87 |
> |
88 |
> My best recommendation is to separate swap and filesystem devices. |
89 |
> While I didn't do it that way, I still separate them through bcache |
90 |
> and thus decouple fs access and swap access although they are on the |
91 |
> same physical devices. My bcache is big enough that most accesses |
92 |
> would go to the SSD only. I enabled write-back to have that effect |
93 |
> also for write access. |
94 |
> |
95 |
> If you cannot physically split swap from fs, a tmpfs setup for portage |
96 |
> may not be recommended (except you have a lot of memory, like 16GB or |
97 |
> above). But YMMV. |
98 |
> |
99 |
> Still, I recommend it for /tmp, especially if your system is on SSD. |
100 |
> Unix semantics suggest that /tmp is not expected to survive reboots |
101 |
> anyways (in contrast, /var/tmp is expected to survive reboots), so |
102 |
> tmpfs is a logical consequence to use for /tmp. |
103 |
|
104 |
|
105 |
-- |
106 |
Regards, |
107 |
Kai |
108 |
|
109 |
Replies to list-only preferred. |