Hi Bryan,
> I have a question. A client of mine requires a highly scalable HA web
> server. The total volume of php scripts and html for the site are minimal ( <
> 2 MB ) I am considering proposing an lvs-dr setup where each node is
> diskless. The director would be diskfull and a hot-backup installed. The
> nodes would boot off their NICs, download a kernel image and mount an NFS
> root filesystem off the director ( or another box if this load is too high ).
> The node then would create a 2 MB ramdrive and rsync the php files and htdocs
> to it. Then they would signal the director they are ready to be added to the
> load balancing pool and receive hits.
Aha, the SUN diskless-client approach. Nobody has tried it before but it
should certainly work if you managed the diskless clients to boot via the
PXE and linux. What for is the NFS mounted share needed? Excuse my ignorance.
> My question is: Is the ramdisk really necessary to acheive optimal
> performance/dollar spent considering Apache's inherent caching mechanisms?
Honestly I don't know about the inherent caching mechanisms but since disks
nowadays are so bloody cheap, you'd better save some nerves of yours and
use some disks. The bootp support for the kernel stripping is a pain in the
ass under linux. It's not anywhere close to for example SUN's JumpStart.
> Secondly, diskless motherboards are seemingly very hard to find. In an ideal
> world I would find a 200 MHz FSB (dual would be amazing) Socket A with
> onboard 2 port bootable NIC and enough room for 1GB ram. Does such a beast
> exist out there? Or anything close?
Let me propose you a different approach: Take a Intel L440GX+ board and
check out the linuxbios project: http://www.acl.lanl.gov/linuxbios/index.html
Burn linux onto your nvram and have a fully featured Linux system within
seconds. It's working!
Best regards,
Roberto Nibali, ratz
--
mailto: `echo NrOatSz@xxxxxxxxx | sed 's/[NOSPAM]//g'`
|