On Thu, Jan 22, 2009 at 03:39:54PM +0100, Attila Nagy wrote: > Hello, > > In /etc/rc.initdiskless there is a function, which creates memory disks in > diskless environments: > # Create a generic memory disk > # > mount_md() { > /sbin/mdmfs -S -i 4096 -s $1 -M md $2 > } > > I have a lot of remote booted diskless and "with disks" machines, which > rely on this kind of storage. The problem is that the above command > specifies "-M", so it will create MD_MALLOC disks, which can't be swapped > out, so it constantly takes away RAM, even if there is only a lightly used > dataset on the storage, which could be in swap too in cases, when there is > a memory pressure on the system. > > So the question is: what is the rationale behind creating malloc backed > disks by default, instead of swap-backed ones? > I can only think of two: > - MD_SWAP disks cannot be created, if NO_SWAPPING is enabled in the kernel > (I haven't checked, if the swap code is enabled (default) and there is no > swap, I can create swap based disks, like malloc based ones) > - under memory pressure, the swap based disks will be slow, so maybe it's > not a goot idea to put /etc (in netbooted environment, this is by default > on memory disks) onto it. BTW, I don't see the difference here between a > netbooted machine, having /etc on a swap backed memory disk, which also > holds swap and a locally booted machine, having /etc on a disk, which also > holds swap. (of course there is a difference, if the swap is on another > disk(s) > > So, are there any objections on changing > /sbin/mdmfs -S -i 4096 -s $1 -M md $2 > to > /sbin/mdmfs -S -i 4096 -s $1 md $2 > > ? It's a historical artifact rooted in the misleading name of the swap-backed type. I'm having a hard time imagining a case were it makes any differece and you'd actually use the script, but we should generally use swap backed mds. -- Brooks
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:40 UTC