I just wanted to ask what's the position of Cisco in the new vSphere 5 licensing model (per vRAM). As far as I understand, that will imply a big kick in Extended Memory's value. What are your thoughts?
this vRAM licensing model will KILL VMWares market share…i know a LOT of (formerly loyal) VMWare customers are looking at other hypervisors now.
just do the math:
B250, 2 sockets and 384 GB RAM:
ESX4: 2 licenses
ESX5: at least 8 licenses
B230, 2 sockets and 512GB RAM:
ESX4: 2 licenses
ESX5: at least 11 licenses
so we are looking at 4-6 times the licensing fees (plus service of course) and now also moores law works against you… newer server with more RAM - buy even more licenses.
for ONE 1TB VM you will need 22 (enterprise) licenses!!! and don't even start thinking about desktop virtualization… it is a lot cheaper buying hardware! (or using XEN of course!)
(at least because you will only get 48GB with the most expensive licenses… the smaller editions have less, the free ESXi5 is limited to 8GB per socket)
i guess Microsoft and Citrix are quite happy now…
it's shame, i really like VMWare and i think they are the superior platform… but with that licensing model i cannot recommend VMWare to any customer any more
just check the threads in the VMWare forums...
Here's my take on this change.
Within the last decade Datacenters went through a similar transition too changing their billing method from Rack Space to the current Power Usage model. In my first DCs I would ask for a Rack, and was given enough power and connectivity to support as many servers in that rack as I could fit. Well, with the advent of blade enclosures and 1U rack servers we went from being able cram in 4-5 servers, up to 20+. This obviously didn't fair well for DC cooling or their revenue's sake.
VMware is at a similar point where people are cramming more & more VMs on each host, and the socket counts are not increasing (as per Intel's direction). Personally, I think a more appropriate move would have been to move to a core count licensing model, but there might be issues with hyperthreading vs actual core count, but I'm sure this could have been worked out. Back to the issue, VMware being the market leader has the luxury of "calling the shots" at this stage in the game. If people want to rip and replace, that's going to be a very costly change - and VMware are aware of this.
In terms of Cisco, we're sticking with our current licensing model (in terms of the N1K & socket count) for now. I haven't heard any definate plans for us to follow suit with VMware as acknowledge the same concerns as others. That being said, we're also partnering with other Hypervisor vendors to provide customers with options. I've been a VMware supporter for over a decade and I see this as a natural change to the game.
It would be great if VMware licensed to their customer's concerns rather than blind-siding them with such a costly licensing model change like this. Do I think VMware will change their model - no. Will Microsoft & Citrix capitalize on this move - definately. This might just breath some healthy competition into the Hypervisor market which was needed.
don't blame VMware for changing their licensing model but they way they are doing it is crazy. The should offer "vRAM" expansion licenses. Why should we have to buy a complete license if all we need is more RAM?
Sent from Cisco Technical Support iPhone App
That echoes a lot of what's being said on VMware's forums. Their reasoning is sound, but the implementation was shortbrained. 48GB per socket for Enterprise Plus?! Who honestly thinks that's enough for a license that yesterday had no limits?
Sure 95% of customers' "current" needs won't be affected, but who bought and budgeted for their VM licensing based on moment in time needs? I purchased a UCS and VMware infrastructure with a 5year plan in mind, and by my growth calculations I'll need to add 41% to my VMware licenses after only 2years. I'll be at 93% of my vRAM by the end of this year. It's not in the budget, so when we get there hopefully Hyper-V or Xen has caught up sufficiently. We already buy MS Datacenter for our hosts to support the guests. SCVMM is cheap by comparison.
So yes. I'd love to see vRam packs as well, but I'd also like to see the initial allocations at least doubled and/or see people with SnS given as many licenses as they currently have physical RAM.
I've already sent my views up through my Cisco and VMware reps. If you haven't already please do that. VMware is getting the message, but the more they hear the better the outcome for all of us UCS customers.
Herald, I totally agree. I literally cut a PO for VMware 4.1 and less than a week after I hear about this new structure. I capitalized on the per processor license and crammed 128 Gb of ram per server with my environment run on 4 servers. Now my vmware environment is licensed for 384 Gb of ram, meaning I should have saved my money and not purchased the 4th server because I can't use it.
I am a local government and we are struggling right now, I don't have the budget to increase my VRAM entitlements!
I haven't yet talked to my account manager about this but trust me I will be having a talk... Maybe we should post our account managers emails and everyone email all account managers to complain! Ha, maybe then they'll get the point.
I believe this change is driven from greed, but as you said those of us with enterprise plus will not purchase 48 Gb per server, we need greater density!
Sent from Cisco Technical Support iPad App
Doh!..I should have waited a bit to respond. Looks like VMware heard, listened, and reacted appropriately.
The amount of vRAM that counts against your vRAM license pool will be capped to 96GB per VM !!! In other words, even if you assign 256GB or the new 1TB limit of RAM to a VM, it will only count as 96GB for your license."
While I am not a fan of vRAM, it really may not be as bad as many people are making out to be. For starters, many folks have interpreted some concepts incorrectly. It's not the amount of RAM that you have installed on the host that counts, it the actual amount of RAM that is allocated and in use that is being measured. It will be interesting to see what happens with marketing regarding the use of overcommit and TPS. The two technologies were promoted in the early years as a way of loading up a host, but the net effect may result in higher costs for customers. However, current best practices is to not load up a host for failover reasons. If you keep a percentage of memory in reserve for failover events, that memory is not "taxed".
Second, I think the driver for this is that dreaded word "cloud". It seems to me that VMware is moving licensing towards a consumption model vs an infastructure model in order to facilitate the move to the cloud in the future. If your organization allows for chargeback, vRAM is the way to go because you can now tie usage costs to workloads. Of course, to do that you will need some sort of mechanism such as VMware's Chargeback (at an additional cost I might add). I only foresee this being palatable if the hardware vendors adopt something similar since we currently have to purchase the infrastructure up front.
@Harry - I hear a lot of discussion on Twitter regarding vRAM expansion packs. Don't know if it's true or not, but the folks discussing it tend to be correct more often than not.
Another item for thought - there is no hard enforcement mechanism for vRAM. I'm guessing that this is VMware's way of providing a grace period and giving people time to get used to the idea.
One last thought - I agree with what is being said about contacting your rep. Let VMware know your thoughts on the new licensing model. Who knows, VMware may change course.
I'm definitely anxious to see if that Derek Seaman rumor proves to be true. That would indeed quell 80+% of the complaints. The reaction on VMware's forum seems mixed.
As for my $0.02, I think generally speaking the complaints are taking 3 forms.
For 1 and 2 I think this new rumored proposal satisfies all but the highest-end systems today and people who were heavily using overcommit and/or running big hosted or VDI deployments without using the View license model. There were a number of folks who had existing Enterprise and Ent+ licenses who used them plus a 3rd party front end for VDI. They will still anger a bunch of their biggest customers, but those will likely negotiate something on a a one by one basis.
For those complaining about 3, I think this is an interim step on the way to a more cloud-compatible model. A reasonably priced vRAM expansion pack would be a logical next step in that direction. Eliminating the per-socket requirement would be another step. They need to just quit the marketing spin and fess up that this is a compromise step in the cloud direction, not the wonderful panacea of beautiful licensing perfection they initially made it out to be.
I reached out to my AM at VMware asking about this rumor. Here is their response. I think the rumors are ture or very close...
"VMware has been listening to customers, and is planning to make an announcement next week. I cannot comment at this point."
For anyone not watching the VMware channels:
This is a step in the right direction. It doesn't cover those with fully booked farms of fully populated B250s or B230 M2s, but it looks like they played the 80/20 rule and did what they could to cover 80% of the uproar. I applaud them for responding pretty quickly.
I still think they're going to have to offer vRAM expansion packs eventually, but this fills the gap until they can move to an even more cloud-centric model.