[Leaplist] Power mitigation, diskless v. truly thin, virtualization and/or remote -- WAS: Scanner Setup Help?

[Bryan J. Smith]

Richard F. Ostrow Jr. <rich at warfaresdl.com> wrote:
> A major version of "diskless" is that you spend real money getting
> a nice, highly redundant file server, and the client machines
> benefit from the redundancy built into the file server - ie, they
> don't need expensive RAID cards or anything similar. Of course,
> doing this for a desktop system probably isn't a good idea...
> 100MBit ethernet will only get you 10-12 MBytes/sec, whereas a
> local hard drive will get you a 10X performance increase.

First off, it depends on the implementation of diskless.  If you
are using read-only mounts with caching (various filesystem
operations) and tmpfs for coalesced writes, then you can mitigate
performance issues.  In some cases, you may see other benefits,
including security.

Secondly, this brings me back to the fact that diskless != thin
client.  Thin clients don't access binaries, libraries, etc...
over a network, but only display remotely running programs.  So
that removes the entire network bottleneck, as long as the remote
display is using an efficient widget/meta communication.

The only issue with thin clients remaining are bitmapped operations.
They eat both bandwidth and unnecessarily taxi the server with overhead.
E.g., streaming video in a browser window.  Red Hat's SPICE protocol
was designed to remove this issue, making thin clients a desktop
reality without common Internet browsing issues.

> Personally, I utilize a master RAID server with real
> redundancy hardware (poweredge 1650 with 3 1TB SATA drives
> connected to a RAID-1 with 256M battery-backed RAM, as well as
> 768M ECC registered RAM, lights-out management, etc... all
> for ~$715) ($150 server, $75 lights-out management
> module, $80/drive, $250 RAID card). One drive is configured
> as a "hot spare", meaning it will automatically reconstruct
> the array in the event of a failure (buying me some time to fix
> it before another drive fails, which DOES and HAS happened 
> killed my last array)).

Yes, hot spares should be considered mandatory with today's disk
failure rates in excess of 2%/year (400,000 hours MTBF).

> Note that as of today, you can get 1.5 TB drives for $90 each.

Indeed.  And 2W (tiny fraction of a watt when idle) 2.5" drives
are running $75 for 500GB, $50 for 320GB.

> To this, I connect my mythtv backend, which boots off of
> NFS to take advantage of the redundancy...

What about your network redundancy?
Also, have you considered local CompactFlash instead?

> and then tosses it out the window by using a software RAID-0
> for the recordings across 3 1TB hard drives (so the system is
> fault-tolerant, but the recordings are highly susceptible).
> Unfortunately, my experience thus far has been that linux
> only barely supports diskless systems, and frequently breaks
> that support for lengthy periods of time with kernel updates
> and baselayout updates.

All Fedora and Red Hat Enterprise Linux releases have supported
"Stateless" for a long period of time.  I have implemented such
for both diskless and thin client.

It only requires identification of select directories for write
operations.  Most of the time, the defaults "just work."

In the case of Thin Clients, the Thin Client only needs a minimal
environment (X, maybe ICAClient, RDP, etc..., assuming you're
doing a full, remote X root -- otherwise you need a window manager
as well), which completely simplifies the issue.  I.e., /tmp
and a few other things.

> For example, a recent kernel update removed TCP packet support
> from NFS... which causes problems on my switch (admittedly, not
> a linux problem, but a switch problem... but using TCP packets
> fixed it).

Did you mean UDP?  I'm not following.

> Personally, for low-power systems (of which I have three),
> I use solid state over CF cards with a Via architecture
> (usually C7 CPUs). These have high temperature tolerances
> and very low power consumption... they've been running in
> my garage without A/C or ventilation for 1.5 months thus far.

Before the Intel Atom, the long-standing Cyrix, National Semi
and Via products from Geode through C3 through C7 were the lowest
power x86 solutions.

> Basically, these are simple data processing machines
> (email, web server, proxy (purely in RAM, not in flash),
> bind9 server, etc) that don't need tremendous amounts of
> storage... or if they do, they can tie in over NFS
> to my file server.

Candidates for virtual machines.  For Linux on Linux, Xen paravirt
is still a very capable and high performing solution, putting KVM
and fullvirt aside.

-- 
Bryan J Smith          Professional, Technical Annoyance
b.j.smith at ieee.org    http://www.linkedin.com/in/bjsmith
--------------------------------------------------------
I don't have a "favorite Linux distro."  I use, develop
and support community efforts, often built around Linux.
Technology and solutions are my focus, not dragging in
assumptions, marketing and other concepts which dominate
non-community developed software, which I left long ago.




-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.