build a kernel over a mounted NFS partition

[Valdis.Kletnieks at vt.edu]

On Mon, 03 Nov 2014 21:00:13 +0200, Kevin Wilson said:

> The strong server, on which I intend to build, has about 4 time more
> CPU power than the NFS server.

Actually, CPU doesn't matter as much as I/O capability.  You don't believe
me, try building a kernel with a cache-cold source tree, and again with
a cache-hot tree.

A long time ago in a $DAYJOB far far away, I was the admin for a small
cluster of Sun-3 workstations.  A Sun-3/50 had a whole whopping 4M of
RAM, and an optional attachment for a very slow "shoebox" disk drive.
It was actually faster to put the swap space on a file on the NFS server (a
Sun 3/280 with 16M of memory and Fujitsu Super-Eagle disk drives) and
swap across the 10mbit ethernet than to the local drive.

> The question is - does it worth it ? or building by NFS is too much
> expensive because of the overhead ?

Depends on your network and the NFS server, and the disk system that the
NFS server uses as a back end.  My NFS servers usually have a 10G ethernet
card, and expensive high-end disk systems(*) with several hundred disks to
stripe RAID across.  Usually pretty hard for a system that has a commodity
local hard drive to compete with that sort of thing.

Only way to tell for sure is to actually try it.

(*) Stuff like the DDN SFA12K http://www.ddn.com/products/storage-platform-sfa12kx/
or the NetApp E-series http://www.netapp.com/us/products/storage-systems/e5400/e5400-product-comparison.aspx


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 848 bytes
Desc: not available
Url : http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20141103/c9a305fb/attachment.bin