| 1 |
On 6/16/19 1:14 PM, Wols Lists wrote: |
| 2 |
> I'd have a single /home partition |
| 3 |
|
| 4 |
I was thinking of the other OS as more of a live distro copied to the |
| 5 |
system than anything else. I wasn't thinking that the OP wanted to |
| 6 |
actively use the alternate distro frequently. As such, I figure that |
| 7 |
most customizations can live on the main OS and it's associated home. |
| 8 |
|
| 9 |
> Drives are cheap. The old "swap is twice ram" rule actually isn't an |
| 10 |
> old wife's tale - the basic Unix swap mechanism NEEDS twice ram. |
| 11 |
|
| 12 |
No, it doesn't. Not any more. It hasn't for quite a while. |
| 13 |
|
| 14 |
Swap was FAR more important when there wasn't enough ram for the |
| 15 |
server's workload. Or when the workload was transient like a multi-user |
| 16 |
system. (Think terminals and / or telnet and / or ssh sessions for many |
| 17 |
users logged in and sporadically using the system.) |
| 18 |
|
| 19 |
Red Hat's recommendation last I looked was the following: |
| 20 |
|
| 21 |
If the system has ≤ 2 GB of memory, have 4 GB of swap (if you can). |
| 22 |
If the system has > 2 GB of memory, have the same amount of swap as memory. |
| 23 |
If the system has > 16 GB of memory, have 16 GB of swap. |
| 24 |
|
| 25 |
Take a look at the output of free on most systems. I'm betting that you |
| 26 |
won't find very much swap used, if any. So dedicating 64 GB to swap on |
| 27 |
a machine with 32 GB of memory is … silly. Especially if you never have |
| 28 |
more than about 100 MB ~ 1 GB of swap used (if that). |
| 29 |
|
| 30 |
You can probably get away with < 1 GB of swap on many systems. But |
| 31 |
there is a different thing where that small amount of swap starts to be |
| 32 |
an issue. That's when you want to do things like take a dump of kernel |
| 33 |
memory and the stack. That does need some space. But I think 1 or 2 GB |
| 34 |
is plenty. |
| 35 |
|
| 36 |
> Okay, optimisations turned "must" into "should", and the swap mechanism |
| 37 |
> was seriously revamped many moons ago and may have changed things |
| 38 |
> completely (I've never managed to get anyone knowledgeable to tell |
| 39 |
> me what happened), but what I do is always ... |
| 40 |
|
| 41 |
We've also drastically changed how we use Unix systems. We no longer |
| 42 |
have 25 ~ 250 people logged into them via terminals. Now most Unix |
| 43 |
systems are dedicated to a single task, be it web serving, or a |
| 44 |
database, or something else. |
| 45 |
|
| 46 |
Plus, we don't want those workloads to be running in swap, so we give |
| 47 |
the servers more memory thus making them even less likely to hit swap. |
| 48 |
|
| 49 |
> Multiply my mobo's *maximum* ram by two. For *each* disk, create a swap |
| 50 |
> partition that size. Add all swap partitions in with equal priority. |
| 51 |
|
| 52 |
I think that's bad advice and I discourage that. Especially if you're |
| 53 |
running all SSDs and your system can take half a TB of memory. Do you |
| 54 |
/really/ want to dedicate 1 TB of each SSD to swap? Just how big are |
| 55 |
the SSDs anyway? ;-) Also, if you've got eight or more SSDs, your |
| 56 |
recommendation would mean that you have 16 times the memory as swap. It |
| 57 |
would be even worse on a server with 24 x 2.5" SSDs. That would be 48 |
| 58 |
times the memory. |
| 59 |
|
| 60 |
> It has been pointed out that this is not necessarily a good idea, |
| 61 |
> a fork bomb would cause havoc because the machine would grind to a |
| 62 |
> swap halt long before the OOM killer realised anything was wrong, for |
| 63 |
> example, but it suits me especially as I put /tmp and /var/tmp/portage |
| 64 |
> on tmpfs. |
| 65 |
|
| 66 |
To each his / her own. |
Archives