On 1/23/10 6:16 PM, "Patrick Spinler" <[email protected]> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> David Boyes wrote:
>> KVM is pretty immature in terms of actual usable management tools.
> Even in the commercial space?

Yes. You'll find that most of the Linux management tools (even the
Intel-only ones) don't understand what to do with KVM, and the few that do
recognize an active KVM infrastructure have only minimal function. The idea
of collecting console information for KVM virtual machines for later
analysis, for example, is missing from all the ones I'm aware of. Basic
stuff that you would expect to exist, not even the advanced stuff like
cloning or copy-on-write or shared R/O disk, or even multiple network
interfaces (example: IBM Director's automated deployment gadget chokes if
you need to define more than one network interface, even if you aren't using
KVM). 

They also scale VERY poorly -- a few dozen instances tend to tax most of
them -- a few hundred and you're looking at multiple copies of the tools. A
few thousand -- they run in fear. Automation infrastructure is weak (if it
exists at all), and any abilities to manage by exception take a lot of human
thought to implement.

There are some interesting system management tools developing in Europe as
part of the CERN ELF management infrastructure, but they're pretty immature
too. At least they're designing for a realistic number of machines, though
-- I keep telling HP and Tivoli that 500 systems per copy is a ludicrous
design point for "enterprise" systems -- come talk to me when one instance
can serve 15,000 systems reliably and then we'll talk manageability and
integration problems across administrative domains and role-based access to
subsets of the infrastructure. CERN's design target is 30,000 machines per
instance, which is affordable and something that a real enterprise might
consider deploying.

> I note there's lots and lots of people calling me with commercial
> products wanting to manage my linux (tivoli, director, satellite, ECM,
> ...)  Surely some of them have to be decent.

Optimist. 8-)

I was specifically referring to KVM support, but in general, they're decent
if you don't try to force them to scale or to manage configurations more
complex than one box, one network, one disk. KVM support is also so new that
what's there is mostly check-box (do we choke when KVM is active? No.
Check.) support that doesn't actually DO anything.

Maybe I'm an extreme case, but I get to touch a lot of these systems. I'd
love to sit the designers of some of these tools down in front of a
Netview/390 deployment in an active sysplex and watch their heads explode at
the message rates that software is capable of. Or ATT's G3 switch management
infrastructure. Or just ONE of CERN's 15-20 large computational cluster
environments. As this cloud computing thing develops, these aren't going to
be outliers -- that's what we can expect in general deployment, and
virtualization aside, the current generation of management tools are going
to need serious reengineering to cope.

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to