[mythtv-users] Diskless

[Chris Rouch]

On 1/31/07, Rich West <Rich.West at wesmo.com> wrote:
> Andrew Wilson wrote:
> > I run a diskless frontend on an epia m10k with 512MB RAM. It works
> > just fine. Root filesystem is on the network. The server is a low
> > power Maxtor Shared Storage - similar in concept to Linksys NSLU2. It
> > boots the frontend using the excellent dnsmasq server for dhcp and
> > hpa-tftpd for tftpboot - took a while to get this going, but now it is
> > flawless. The frontend power button is ACPI'd to suspend to RAM, so it
> > 'boots' in about 5s unless I need to power off. Frontend draws about
> > 3W when suspended - which is about the same as when it is OFF but
> > still connected to the PSU... Shocking!
> >
> > The backend is a cheap (£20!) compaq small form factor office pc which
> > lives in a cupboard. It has 256MB RAM a pair of (cheap) DVB-T cards,
> > and a small disk for OS and swap, though I've not needed the swap and
> > I could probably boot this over network as well. Recordings are stored
> > on a NFS mount on the MSS.
> >
> > The Maxtor box runs cool, draws about 7W and is on 24/7. It also has
> > the task of waking the backend when a recording is imminent. The
> > backend draws about 25W when on, and the frontend draws about 30W. I
> > also have a belkin wireless adsl router which draws about 17W. In
> > theory the MSS could do this job as well, but to be honest I've not
> > tried, and I don't really have time.
> >
> > Everything works OK. The network is wired 100 base tx, not even close
> > to being saturated. When HD-DVB comes along I'll probably need to
> > upgrade the frontend, but hey I'll need to upgrade my TV first! The
> > network should still be OK, and the backend should be fine as all it's
> > doing is copying a still fairly low bitrate data stream from a PCI
> > card to Ethernet.
> >
> > One problem I have is that it seems to be impossible to do wake on lan
> > from suspended (S3) on the backend, probably due to 20th century
> > hardware. So the backend does a normal shutdown when it's idle.Still
> > it boots up mysqld and mythbackend in about 30s which isn't bad. It's
> > just a shame that the frontend is ready before it.
> >
> > What would be REALLY nice is to run a kind of boss backend on the MSS
> > which knows nothing about recording TV, but can serve mythweb and
> > recordings to the frontend on demand. Then I would not have to wait
> > for the real backend to wake up. This should surely be possible in
> > 32MB RAM, though maybe a full MySQL database is too heavy for this
> > task. Perhaps I could do some sort UPnP thing for this, and not bother
> > with a myth frontend. I do all scheduling though Mythweb anyway, and
> > the frontend is only used for viewing recodings.
> >
> > Hope this helps you see what's possible,.....
>
> Wow, definitely.  I'm amazed at just how many people do this, and the
> varying methods in which they use it.  Actually, I shouldn't be so
> surprised with the flexibility of MythTV and Linux, but still... :)
>
> I have taken a dump of the root filesystem of the only FE that I can
> tinker with without having my family lynch me and placed it in an
> NFS-exported area on my main file server.  I've also tweaked the tftp
> config so as to have a file for this particular MAC address.
>
> I haven't begun to test anything yet, but a couple of questions come to
> mind:
> How accurate is the wiki on this topic?
> What, if anything, needs to be done to the NFS root volume's /etc/fstab?
>

If you're running FC4 then system-config-netboot will take care of all
the tinkering for you.  I've yet to get it working for FC6 and I never
used FC5.

I think the wiki is accurate (for fedora at least), except that you
might need "ramdisk_blocksize=1024" in the append string for the
netboot kernel for FC6.

I'm using a VIA EPIA M10K, with 512Mb ram, and xvmc enabled. I only
see problems when the nfs server is overwhelmed with something else.

One problem with my via-rhine network chipset is that if it was
shutdown properly in FC4 then the network wouldn't start again, and so
obviously net booting wouldn't work. I'd have to boot from a disk and
then kill the power before the disks mounted. I patched the kernel to
prevent it reaching this state. Newer kernels accept a command line
parameter "via-rhine.avoid_D3=1" to prevent this.
(linux-2.6.18/drivers/net/via-rhine.c is well commented on this). Once
I get an FC6 net boot working, I'll find out how well this works.

I found the noise from the cpu fan on this board to be too loud, so
(following the hint someone posted on here) I replaced it with a
zalman northbridge heatsink. And as there's no disk inside I've turned
the case fans down to min speed and to my (admittedly worse for wear)
ears the box is as good as silent.

Regards,

Chris