On 27/11/17 04:08, John Andersen wrote:
I've never been more pleased with the performance of this old machine.
Soon as I get it to 4.14 I'll be happier yet, because swap is gaining a multiple pointer system into swap space, one for each core. See https://lwn.net/Articles/704478/ That means swap is not going to be a bottle neck any more.
The case for encouraging swap use is getting STRONGER, not weaker. https://lwn.net/Articles/690079/
If you've got multiple drives, provision swap on EVERY drive, then in fstab set "pri=x" to the same x for all of them. That will then create a *striped* array for swap, improving performance. If you don't specify priority, linux creates a linear array, that uses one drive until it fills up then goes on to the next. Striped gives you speed, but if a disk breaks it will take out every process that's been swapped. Linear will only take out processes swapped to that disk. That said, on my new system I'm going to set swap up as raid-10 with mdadm, which gives the benefits of both at the expense of a bit of disk space.
Don't be stingy when provisioning swap space on an SSD. SSD's need to be over provisioned to protect device life cycle. Got 8 gig of ram? 16 gig of swap is not too much.
As far as I'm concerned the old rule of "no swap or at least twice ram" still applies. Everybody said the "twice swap" rule was an old wive's tale, even before linux was born, then kernel 2.4.0 hit and people found out the hard way it wasn't. I don't know whether the fundamental algorithm has changed since then (that was the trigger for a major rewrite of the swap subsystem), but hey disk is cheap so I take the *maximum* ram my mobo supports, double it, and provision that *per disk*. So I currently have 64GB swap (16GB ram times two times two disks). That's going up to 256GB on my new system (64GB max ram times two times three disks times raid-10) when I get it up and running. (Admittedly, that system is gentoo, so I mount /var/tmp/portage as tmpfs and that needs about 20GB for things like LO, GCC etc.) NB - If you've got a small SSD, set "pri=1" for the SSD, and "pri=2" (or is it the other way round? Look it up) for the HDs, and linux will preferentially use the SSD, only using rust when the SSD is full. Cheers, Wol -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org