[ale] Virtualization
Jeff Hubbs
jeffrey.hubbs at gmail.com
Wed Feb 25 09:58:19 EST 2009
My experience - by which I mean VMware ESX in two workplaces and Xen on
my own - is that virtualization is a case of making deals with the Devil.
Hardcore this-is-what-I-run-production-on ESX use calls for exquisite
levels of care, feeding, money, and intelligent thought. If your IT
organization doesn't have enough of any of these elements to expend on
your ESX plant, epic FAIL awaits. Everything at every level must work
well and perfectly, for they will be stressed and anything that's not up
to the task will have everything on and above it crashing down.
VMware did a reasonable thing by making ESX something that you install
on bare metal, but in exchange you lose interoperability and
flexibility. I liked dealing with Xen (at least as it existed in
mid-'07) because it insinuated itself at a fairly low level - your host
instance can be truly yours - and I am drawn to the idea of using
virtualization for R&D purposes, i.e., that's where server instances
"gestate" before going to real hardware for testing and then production.
I am also seeing virtualization as a legitimate way of addressing
spiraling electrical power consumption. It makes me feel as though
using virtualization to *some* degree could be reasonably considered to
be a requirement for responsible computing.
Disk subsystems seem to be an Achilles' heel of VM rigs...the next time
I try anything like this, I want to try basing it on ATA-over-Ethernet.
The high-dollar canned (rhymes with "crap") iSCSI subsystem in use at a
former workplace of mine was way too accident-prone to depend on.
- Jeff
Jim Kinney wrote:
> On Tue, Feb 24, 2009 at 9:50 PM, Christopher Fowler
> <cfowler at outpostsentinel.com> wrote:
>
>
>> This is my experience as well. I think many companies look at
>> virtualization as a way to save
>> money on hardware.
>>
>
> Virtualization seems like a cool toy. But when I see a business use
> many virtualized machines for daily processes, mission critical
> services, etc, it just screams "single point of failure with massive
> consequences".
>
> It also speaks volumes about the overall architecture and design of
> the processes in use that they require multiple machines for load that
> then get virtualized to save money on hardware.
>
> ?!?!?!?!?
>
> Huh?!? WHA?!?!?
>
> Picture this scenario: Product FOO is composed of database, app logic,
> and UI frontend. The designers all insist that their portion requires
> an independent machine to avoid resource conflicts. So 3 VMs get built
> thus placing all the parts on the same machine with even higher
> overhead than if they were on a single, physical machine. Management
> viewpoint is they don't have a new chunk of hardware to buy for this
> process. While true, they did have to buy a HONKIN' box(s) for the vm
> server.
>
> It always seemed to me that virtualization is a good thing for test
> environments and extremely light loads that are not mission critical.
> But the ideal use in a mission critical environment is as a backup
> environment for the real hardware.
>
More information about the Ale
mailing list