VMWare Rolls Out Their Largest Product Release

opieum writes "VMware has launched Virtual Infrastructure 3.0 today which includes ESX 3.0 and a number of management utilities." Relatedly Jane Walker writes "SearchOpenSource has two authors that try to show why VMware ESX Server is miles ahead of Xen and Virtual Server. Discover what to watch out for when running ESX Server and how to avoid sprawl in your virtual data center."

Comments

Re:VMWare Server 1.0 same as VMWare Workstation 5.

[ruckc]

Then was their GSX Server equal to their workstation product? And if so why are they charging $189 for a workstation product when Server now does the same thing?

Re: So?

[L.Bob.Rife]

As it happens, I can't always get by in my job using only free non-commercial software. Now, I have to assume that several other people here are in the same boat, and commercial software can provide value to them. Given those circumstances, I'd prefer seeing a debate about the relative merits of particular software packages, and discuss it, rather than dismissing a product because it costs money. And if slashdot happens to make a side profit, more power to them.

Benchmarks are too easily rigged.

[jd]

There are some excellent profiling products around (VTune, PAPI, DAKOTA, KOJAK, and those are just the ones I've used). Companies like VMWare probably use some form of profiling already - they'd be insane not to, as it's a great way to improve performance with little effort. Obviously, they'd be equal idiots if they published all of the stats churned out, but there are likely ways they could publish a set of general indexes or tables to show the overheads of running N OS' over M processors with P cores each, plus the cost of running some of the standard administrative functions. Because I'm talking about the low-level operation, rather than custom-made scenarios, the figures won't represent any given scenario exactly but can't be rigged by selecting a given example either. If other vendors then wanted to publish their own figures for the same matrix and functions, then people would have something to work with on comparisons.

Sure, you can probably plug the numbers into a suitably complicated equation, but it won't be linear and it won't be "obvious". The maxima won't be at the same place for different hypervisors, either. That's the point. If you use a single number benchmark, you can (almost) always find something product X does better than product Y. If you have the full behaviour of the system written out, vendors can't obscure things like that. It's good for the customer, as they can then see what product does the best with the specific characteristics they have in mind. It's also good for the vendor, because there's no pretense and no FUD (so the customers like you) and there's no denial (so the developers respect you).

Now, are ANY vendors going to do this? And I'm including Xen and VServers in this. Probably not. There are risks involved in being that transparent, plus costs. And even if the vendors all agreed it was a good idea, you think ANY of them would volunteer to go first?

This is not to diss VMWare. I respect them (as much as I respect any corporate entity) and this is just as true of the Open Source solutions. It's merely the practical reality that promoting a product through total education of the consumer is something neither party really wants. Customers want plausible denial if things don't work out, and vendors are not going to tell you to go to their competitors.

Progress / money ratio

[Anonymous Coward]

This is the real ratio. With this ratio the best virtualizer is by far kqemu. So novell instead of funding Mono and other technologies that are after the microsoft red herrings, please do us a favor and give some money to Fabrice Bellard, to open up kqemu. (which he rightly deserves as he also happens to be the author of FFMPEG).

Re:VMWare ahead of Xen eh?

[ChrisA90278]

When I do any kind of analysis and notice that one company does not want benchmarks published I can only assume that is because they will com out on the bottom and don't want that fact to be published. If they came out on top, of course they'd want everyone to know that. OK now prove me wrong but you will have to publish some benchmarks to prove your case.

But in this case, who cares? No one uses VMs because they need to run something faster. VMs have become popular because we have an excess of hardware performance and can therefor consolidate. Or maybe we use VMs as a test tool or whatever but never for performance senitive applications.

But still, it should be clear to anyone why a company would not want bechmarks of it's products published. They can explain their reasons forever but everyone will always suspect the real reaon is that the benchmarks don't look so good.

Very useful!

[shokk]

The cost of electricity 5 years down the line is probably going to play the biggest role in determining how much hardware you can run AND cool. Consolidating systems into VMs makes much more sense because a VM is not using its entire allocation of memory all the time, so it's easier to save on hardware and energy costs by having fewer systems - the VMWare mantra.

The only place I have seen that this is not true, is when you have a large compute farm, where each system is dedicated to running just that one job, which can take tons of memory (GBs) and always hits 100% CPU for thousands of jobs. In that case I would rather run without VMs because each machine is exactly the same as the next one and imaging a new system only takes a few minutes, localization and all.