I watched the blade discussion with interest.
The amount of compute capacity being contemplated is massive -- it's
well beyond the peak resources needs of everything in the server room at
present.
(that is, if we ignore anyone doing optimal Golomb rulers, prime
hunting, RSA numbers, etc, as these are infinite needs of indefinite
scope that suck up whatever you throw up at them)
If your goal is to just consolidate the current workloads in the most
energy efficient way it doesn't make sense to spend a lot of money on an
action that puts even more capacity online. Whatever you make available
to people will in the end get used. [that is, we will often have a
higher percentage of the blades blazing]
(even if you ban or put a limit on the infinite-indefinite stuff)
You don't need to spend $2,500 when $400 to $900 in upgrades to our VM
server would be enough to consolidate everything running right now and
with capacity to spare. As such, I am launching a capital capital
campaign for that:
http://skullspace.ca/wiki/index.php/Vmsrv#Capital_Campaign
Also seeking project funding:
http://www.skullspace.ca/wiki/index.php/Proposed_projects#VM_server_hardware_upgrades
"""
Current upgrade project is to switch to a CPU with VT extensions, which
will improve VM performance, allow for 64bit guest OS, and also make
more guest operating systems available that are currently a no-go with
Virtualbox and no hardware extensions such as OpenBSD and FreeBSD.
"""
There's a stronger case for power use ROI here -- not only because less
money is being spent but total compute capacity is actually going down.
That is, except for those doing infinite-indefinites, we'll be taking
heavy servers offline that are mostly running idle.
------------------------
And now, to the subject line, as there's no interesting debate in the above.
Someday we will grow and not be saying "shit, we need to consolidate and
reduce energy use". We'll be saying, "more power!" and want to add a lot
more capacity.
It is conceivable that we'll be able to have a successful, special
fund-raising drive just for that, and reach a nice target like $2,500.
But, if you're going to spend $2,500, I say spend it on one, super kick
ass server vs the blade approach of scaling out RAM and CPU in parallel.
I have nothing against blades in general -- for many scientific,
engineering, artistic, and business use cases it makes sense to scale
out in the blade way.
Nor am I against mixing blades with virtualization. (such as here:
http://web.archive.org/web/20090204223932/http://get-admin.com/blog/?p=392 )
What I want to argue is that the workloads of hackers in a hackerspace
are better suited to scaling in a vertical direction over a horizontal
one. Call it, /The one grand machine to rule them all/.
Seeing how donors are already in contemplation mode, I feel the need to
challenge the blade advocates to a debate.
I'm not going to have that debate here on the mailing list (which is why
I haven't said *why* it's better for a hackerspace to spend $2500 to
scale vertically) -- I'm going to give a formal presentation on the subject.
To the blade advocates -- do you wish to accept my challenge to a dual
by scheduling presentations back to back (with random order?)?
Alternatively, I could go first (perhaps late September..) and you could
opt for rebuttal on a separate day once you've seen it?
Mark
_______________________________________________
SkullSpace Discuss Mailing List
Help: http://www.skullspace.ca/wiki/index.php/Mailing_List#Discuss
Archive: https://groups.google.com/group/skullspace-discuss-archive/
