[colug-432] Swap science, was: Install and Manually Partition Debian

[Rob Funk]

On Sunday, March 16, 2014 05:20:22 PM R P Herrold wrote:
> On Sat, 15 Mar 2014, Rob Funk wrote:
> > I don't know what Debian's default is these days, but I
> > generally recommend a big swap combined with /tmp
> 
> Why as to a large swap?

Simply to give plenty of room for tmpfs. Basically I make a swap partition 
instead of a /tmp partition, and size it with that in mind. Anyway, "large" 
for swap is maybe 16-32GB, and is pretty small for a modern hard disk.

>  In the former times a swap 2X
> physical ram 'ruled the roost'.  More recent lore is 'if
> you are into swap', you are paying a high price for r/w
> speeds onto spining rust

Certainly, which is why I keep an eye on my swap usage, and let that guide 
my memory upgrades. But there's also something to be said for letting the 
OS put lesser-used data somewhere slow, to make room in fast memory for 
what's actually being used.

> We have been testing SAN, and adding an invisible cache layer
> of SSD 'just makes sense', save that TANSTAAFL, and that layer
> is not free ;)  Seagate also offers a drive with 64G of front
> end SSD, in front of the spinning platters.  I ordered a
> clutch of these today, for production on an NFS (and
> managed iscsi / LVM) server.  I'll be pulling numbers off to
> see how this affects compile times (I save stats on all
> builds, and will be able to see if / how it helps/ hurts,
> from some log reduction)
> 
> A lot of computer science is invalidating caches well, and the
> 'self learning' tactic which the Seagate device implies may
> work well in my Linux environment, or it may fail
> conspicuously.  We'll see ...

That's mostly about the size of your working set. If the data you're 
working with fits in the cache, it'll help. It it doesn't, it won't help 
much.

-- 
Rob Funk <rfunk at funknet.net>
http://funknet.net/rfunk