Date: Fri, 2 Mar 2018 17:52:32 -0800 (PST) From: "Rodney W. Grimes" <freebsd-rwg@pdx.rh.CN85.dnsmgr.net> To: bob prohaska <fbsd@www.zefox.net> Cc: Ian Lepore <ian@freebsd.org>, "freebsd-arm@freebsd.org" <freebsd-arm@freebsd.org> Subject: Re: Is maximum swap usage tunable? Message-ID: <201803030152.w231qXFA032676@pdx.rh.CN85.dnsmgr.net> In-Reply-To: <20180303014643.GD37148@www.zefox.net>
next in thread | previous in thread | raw e-mail | index | archive | help
> On Fri, Mar 02, 2018 at 05:02:09PM -0800, Rodney W. Grimes wrote: > > > On Fri, Mar 02, 2018 at 10:15:57AM -0700, Ian Lepore wrote: > > > > > > > You forgot a cause: (5) swap is on an sdcard where taking 30-90 seconds > > > > to complete an IO is "normal". > > > > > > > FWIW, the Sandisk Extreme USB flash drive is claimed to be considerably > > > faster than that, ~2MB/sec random write, at least per > > > http://usb.userbenchmark.com/SanDisk-Extreme-USB-30-16GB/Rating/1301 > > > One hopes(!) that the microSD cards of the same name are simlar. > > > > > > > Making it even more fun, there's a sort of (5.5) bullet: an sdcard can > > > > "lend" its horrible performance to every other storage device in the > > > > system. ?If there is a ton of IO queued up to the sd device (such as > > > > when .o files from a a make -j are ending up there) then all the > > > > buffers in the system get stacked up in the sd device's bio queue and > > > > IO to other devices suffers. > > > > > > > A j4 buildworld on an RPI2 running -current seems to work much better > > > than j2 on an RPI3, both using the same Sandisk flash devices. The Pi2 > > > has /usr, /var /tmp and swap on USB flash, the Pi3 has /tmp and half > > > the swap on microSD, with /usr, /var and the other half of swap on > > > USB flash. Thus, /tmp and all of swap are on the same device for the > > > Pi2, while /tmp and only half the swap are on the same device on the Pi3. > > > It was expected the Pi3 should work better than the Pi2, but the opposite > > > seems to be observed. > > > > I would actually expect the opposite, but not for an obvious reason, > > the RPI2 is running a 32 bit version of FreeBSD, and the RPI3 is > > running a 64 bit version. These are memory constrained devices, > > and my theory is that the wasted memory used for pointers that can > > never address more than the 2G of physcial space on the RPI3 are > > killing your memory foot print, and killing your build. > > > Any idea how much memory gets wasted? Right now the Pi3 kernel is > smaller than that of the Pi2, > -r-xr-xr-x 1 root wheel 7142608 Feb 25 17:11 /boot/kernel/kernel > versus > -r-xr-xr-x 1 root wheel 9426388 Feb 26 22:31 /boot/kernel/kernel > on the Pi2. Seems like that ought to at least help... This is an artifact of the number of devices supported being short on the RPi3. > > Getting that 32 bit version of FreeBSD running on the RPI3 would > > prove out my theory. > > > How much and what kind of work is required? >From what jhb said the other day it is minor clean up details, and I believe someone else has wip in progress to clean up some of those details. > > > In my mind it makes 0 sense to run a 64 bit OS on a device that > > physically has less than 4G of memory, other than as an exercise > > in yes, we can make the 64 bit code run on that cpu, but for > > real work usage it makes no practical sense. I also suspect > > that the cortex CPU in that thing can run 32 bit code a wee > > bit faster than 64 bit code other than large data moves... > > > I agree, and remember your observations along those lines some > time ago. The only reason I'm fooling with arm64 is that the older > Pi2 seems NLA and the Pi3 is, or soon will be, the only option. > A Pi3 running Raspbian, on which I'm typing this, is a quite decent > little workstation. It's using USB flash (old and slow, by the way) > for swap, but has a Sandisk Extreme microSD card. And that Raspbian is 32 bit :-D > > > > I've always thought the new(ish) IO scheduler stuff should be able to > > > > help with that in some way, but I never get around to looking at it. > > > > > > > > > > > I'd be pleased to do any testing I can, if it helps. > > > > > Thanks for reading, > > bob prohaska -- Rod Grimes rgrimes@freebsd.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201803030152.w231qXFA032676>