[Beowulf] Clustering VPS servers
landman at scalableinformatics.com
Thu Mar 21 10:11:09 PDT 2013
On 03/21/2013 12:15 PM, Jonathan Aquilina wrote:
> I am venturing back to the cloud side again isn't cloud just a fancy name
> for virtualized servers?
Think of cloud as "hardware and infrastructure as code". You
instantiate what you "need" with some caveats.
Completeness of API, breadth of product is nice, but at the end of the
day, the only thing that really matters, is, does it do what you need at
a price you are willing to pay.
+1 to Chris for message going over these elements, though one has to be
careful as Amazon isn't the only one out there. They are good for some
things, but as many of our customers have discovered by trying them, not
good for everything, nor even the majority of items. Clouds (public and
private) have a definite (and growing) niche, and I see the entire
"hardware and infrastructure as code" as a very good thing. It has a
nasty tendency to create waves of fads though, and its important to be
able to recognize such things.
This said, not having a block, object, or other storage API when you
don't need it really isn't a deal killer. Not having an end-to-end low
latency fabric, and bare metal servers is (for our customers). But our
customers are not necessarily the same as all customers, and the
recursive joke I tell about this is "gross overarching generalizations
tend to be incorrect". Basically being all things to all people means
that none are done well. Focusing upon niches where people get what
they need is IMO a good way to carve out a market.
With all due respect to those whom advocate heavily for one thing or the
other, our experience with this is that the joke is on the end user when
things don't work well at all, and cannot be made to work well due to
design/implementation issues. Especially if they've sold this to
management/board of directors. This is true of internal as well as
external resources. This is more of a generic business problem than a
public cloud vs private resource issue.
This said, a quick pointer over to an IDC paper is an interesting read:
http://www.idc.com/getdoc.jsp?containerId=prUS23972413#.UUs7OqDYSlg is a
good read. Also there are others as well. Public cloud is large and
growing, and largely commiditized. Very non-HPC specific, though with
careful marshalling/selection of resources (CycleComputing) it can be
viable for some HPC. But its very hard to argue a single platform is
all things to all people, when it clearly isn't. Moreover, Amazon
itself just signed a nice sized deal with a government entity for a
private cloud offering ...
Basically, Chris is dead on right with his points on "there must be an
API, and a broad array of services" for a good public cloud. A private
cloud can get more specific on needs, and not worry about implementing
things that are not needed right away.
But cloud isn't always "virtualized". Amazon is largely
para-virtualized which is great to maximize tenancy, but not so great
for performance. Its APIed out the wahzoo. Other clouds are
hyper-virtualized and run closer to the metal. Others still are
bare-metal with provisioning magic atop them.
What likely matters most with public clouds are the ability to move
between them when one of your cloud providers decides to compete with
you. This has been a thread of articles on The Register (which I wrote
about here: http://scalability.org/?p=5898 )
This isn't an Amazon issue as much as it is a business dependency
issue. Freedom to move is important. It would be nice to have a common
API so as to make this work well. The very last thing any of us want is
the Microsoft-ization of the cloud world. That would be, universally,
and catastrophically, bad for all.
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics, Inc.
email: landman at scalableinformatics.com
web : http://scalableinformatics.com
phone: +1 734 786 8423 x121
fax : +1 866 888 3112
cell : +1 734 612 4615
More information about the Beowulf