1 |
On Tue, Sep 4, 2012 at 11:53 AM, Dale <rdalek1967@×××××.com> wrote: |
2 |
> Alan McKinnon wrote: |
3 |
>> On Tue, 04 Sep 2012 09:15:31 -0500 |
4 |
>> Dale <rdalek1967@×××××.com> wrote: |
5 |
>> |
6 |
>>> I think the new method for determining swap is to use what makes sense |
7 |
>>> and not the old rule of 'twice the ram'. |
8 |
>> Alan's new rule of swap is: |
9 |
>> |
10 |
>> If you ever use swap as swap at all, find out how your machine is |
11 |
>> misconfigured. When my 16G is "not enough" anymore, something is badly |
12 |
>> wrong and it isn't not enough RAM and I need swap to wiggle around |
13 |
>> in :-) |
14 |
>> |
15 |
>> I think the 2 x RAM rule stopped being applicable when the average |
16 |
>> machine got to more than 16M. Some old memes are like zombies - very |
17 |
>> hard to kill. |
18 |
>> |
19 |
>> This laptop has a "swap" partition, but it's not for swap, it's for |
20 |
>> hibernate. And I never use it, it takes longer to come out of hibernate |
21 |
>> than to just boot up from cold! These days I just suspend. |
22 |
>> |
23 |
>> None of this changes the fact that the kernel still does get upset when |
24 |
>> it has no swap at all (even just a little bit). But that doesn't mean |
25 |
>> we should still be using it as full-blown swap. |
26 |
>> |
27 |
>> |
28 |
>> |
29 |
> |
30 |
> |
31 |
> Yup. I have swap but I have it set to where it won't use it unless it |
32 |
> is REALLY bad. I have swappiness set to like 20 or something. It will |
33 |
> fill up my ram with cache and such but it rarely uses more than a few |
34 |
> hundred kilobytes of swap. When I see it using that, I usually kill |
35 |
> swap and add it back. I just don't like a machine with 16Gbs of ram |
36 |
> using swap at all. I have thought about setting it to 10. Maybe then |
37 |
> it will leave it alone until it really hits the fan. ;-) |
38 |
|
39 |
Set swappiness to 0. Swap will be used if and only if absolutely necessary. |
40 |
|
41 |
Also, you're unlikely to notice a performance hit if the amount of |
42 |
data in swap is only a few tens of megabytes; the seek-and-read rate |
43 |
of even spinning platter disks should tend to cause that bit of |
44 |
latency to get lost in the normal noise of library linkage, data file |
45 |
loading, etc. (heck, it might even still be in the drive cache) The |
46 |
performance hit is there, but probably not subjectively noticeable. |
47 |
|
48 |
> |
49 |
> That said, I did roll over one night and notice that the CPU was going |
50 |
> ape. I got up and into my chair to notice it was using almost all the |
51 |
> ram and was starting to use a bit of swap. I switched to a console, ran |
52 |
> htop and noticed that some KDE process was using about ~15.5Gbs of ram. |
53 |
> It was crazy to see. I couldn't get it to die with kill -15 so I did a |
54 |
> kill -9. I guess it had to know I really wanted it dead. It has not |
55 |
> happened since so no clue on why it did that. Heck, it ran the same |
56 |
> version of KDE for a good while and still didn't do it. Cosmic rays |
57 |
> from Mars I guess. |
58 |
> |
59 |
> I would recommend at least 500Mbs or so of swap regardless of ram tho. |
60 |
> Some swap is a good idea. Just try not to use it since it is dog slow. |
61 |
|
62 |
Indeed. |
63 |
|
64 |
> If you are using hibernate/suspend thingys then that is different. |
65 |
> Isn't that when it has to be at least as much swap as you have ram? |
66 |
|
67 |
Yes. |
68 |
|
69 |
-- |
70 |
:wq |