1 |
Peter Humphrey posted <200605011141.50417.prh@××××××××××.uk>, excerpted |
2 |
below, on Mon, 01 May 2006 11:41:50 +0100: |
3 |
|
4 |
> On Sunday 30 April 2006 16:51, Mark Knecht wrote: |
5 |
> |
6 |
>> 2) When I built the machine I had 512MB in the machine so a 1GB swap |
7 |
>> seemed fine. I have since updated to 1GB so the swap seemed a bit small |
8 |
>> anyway. |
9 |
> |
10 |
> I've read somewhere that more than 1 GB swap would be wasted. |
11 |
|
12 |
I believe that's no longer correct -- under certain circumstances, anyway. |
13 |
|
14 |
Back with kernel 2.2 IIRC, a swap partition over 128MB was wasted as that |
15 |
was the biggest the kernel would make use of anyway. Further, there was |
16 |
(still is IIRC tho I could be wrong) a limit of I think it was 8 swap |
17 |
partitions (I believe it's 16, now, but without the size limitation, or |
18 |
more specifically, with a size limitation that's well outside reasonable |
19 |
levels -- 2 TB or some such <g>), which @ 128MB a piece indeed total a gig. |
20 |
|
21 |
I know after I set up my 4x4G swap (a bit of boasting... I've worked |
22 |
hard to get this system so I deserve it... 4 SATA drives, each with |
23 |
identical partitions for RAID, with a 4 gig swap partition on each one, |
24 |
all set at the same swap priority so the kernel automatically stripes them |
25 |
for much faster access -- it's /much/ faster than single-disk swap access, |
26 |
especially when the same disk holds everything else and you are doing |
27 |
active I/O on it at the same time it's swapping!), with a gig of memory, I |
28 |
had it using over a gig of swap a few times, and once, when I had a |
29 |
runaway program gobbling RAM, I let it go to about 15 gigs of swap before |
30 |
I killed it before the kernel decided to activate the OOM-killer. |
31 |
|
32 |
As it happens, I've since upgraded to 8 gig of memory and I've never even |
33 |
had that full yet since the upgrade (I've had it up to about 7 gig, |
34 |
including cache and /tmp mounted as a six gig max tmpfs, so the portage |
35 |
temp dir stuff never even touches hard disk -- only memory), so haven't |
36 |
gone into swap at all since then. If and when I do, however, I have a |
37 |
full 16 gig of swap to play with, 24 gig total, mem plus swap, and it's |
38 |
still 4x4 striped, so going that deep into swap isn't going to bring the |
39 |
system to its knees to the point it would on a single disk system. |
40 |
|
41 |
All that said, as I've mentioned a couple times now, on a single disk |
42 |
system, there's a /huge/ cost in terms of access speed, to access swap. |
43 |
Given the cost, practically speaking, it's rather likely one wouldn't |
44 |
/want/ much more than a gig of single-disk swap. To utilize more than |
45 |
that at the raw disk hardware byte per second thruput we are talking... |
46 |
it's simply not all that practical, even if one /does/ need the memory. |
47 |
The disk thrashing and latencies involved just don't lend themselves to a |
48 |
smooth operating system or experience in any way shape or form, no matter |
49 |
how you look at it, if the data's over a gig and the hardware is a single |
50 |
spindle hard drive. Given that, use of more than a gig of swap is going |
51 |
to be more theoretical than practical in any case. |
52 |
|
53 |
OTOH, with the size of today's drives, often a hundred gigabyte or more |
54 |
need not even be partitioned, let alone ever used, and if that's the case, |
55 |
throwing a few gigs at a swap partition, just so the system and system |
56 |
operator have the flexibility to use it if it's ever needed, certainly |
57 |
doesn't hurt. With a dual boot system and a drive < a couple hundred gig, |
58 |
it's possible there's not the room to spare I was talking above, but |
59 |
something reasonable, 2-4 gig, might still be possible, if one wants it. |
60 |
|
61 |
-- |
62 |
Duncan - List replies preferred. No HTML msgs. |
63 |
"Every nonfree program has a lord, a master -- |
64 |
and if you use the program, he is your master." Richard Stallman in |
65 |
http://www.linuxdevcenter.com/pub/a/linux/2004/12/22/rms_interview.html |
66 |
|
67 |
|
68 |
-- |
69 |
gentoo-amd64@g.o mailing list |