1 |
Mark Haney wrote: |
2 |
> Paul Hartman wrote: |
3 |
>> I agree, I use /dev/shm (4gigs) for my portage tmpdir and it has had a |
4 |
>> bigger noticeable speed impact than ccache or niceness, and the |
5 |
>> silence of zero disk activity (other than reading the distfiles in the |
6 |
>> unpack stage and installing the compiled files) is nice, too. |
7 |
>> |
8 |
> |
9 |
> I would do this, however, my problem is (or was) RAM. Until yesterday I |
10 |
> had only 1GB RAM in this laptop. Now I have doubled it which is the max |
11 |
> it will support (it's 4 years old). I don't think 2GB is worth trying a |
12 |
> tmpfs for. |
13 |
> |
14 |
|
15 |
I wouldn't let this deter you. I only have 2GB of RAM in my server and |
16 |
it is doing quite a bit (myth, apache, samba, shorewall, postfix, etc). |
17 |
I just have a nice big 8GB swap partition and I never have issues. |
18 |
|
19 |
In the worst case using lots of tmpfs causes swapping which slows down |
20 |
compilation due to IO. In the best case, not using tmpfs causes lots of |
21 |
writing to disk and slowdown due to IO. In my opinion the worst that |
22 |
can happen with tmpfs is that it just goes no faster than what you see |
23 |
now. In certain well-tuned applications there might be benefits to not |
24 |
using swap at all, but if you have a swapfile now the system will use it |
25 |
whether your RAM is full or not, and at least with tmpfs you have the |
26 |
benefit of separating long-term file storage from short-term file |
27 |
storage and allowing the OS to optimize accordingly. |