Ed Brill

Post a Comment

1. 1  Paul Robichaux http://www.robichaux.net/blog | 7/25/2006 9:33:08 PM

   My thoughts: { Link }{ Link }

1. 2  GarryL  | 7/26/2006 3:44:58 AM

   Sounds very bad, and extremely confusing. Everyone remeber the old Oracle universal power unit (UPU)? What a clever and simple model that was. Not.

   Anyone fancy explaing this to a potential new SME site? Or do you think they will just get confused and use the simple Exchange model?

   Back to the drawing board with this one I think. I prefer the KISS principle myself.

   { Link }

1. 3  NeilT  | 7/26/2006 7:14:20 AM

   This is always a thorny issue, however it always pays to understand where the "Core" issue comes from.

   Ever decreasing performance gains from ever spiralling processor Hz hikes are no longer tenable. So Intel has finally decided (like AMD a long time ago), to scale out instead of up.

   So actually the Core issue is about increasing the power per socket and not about multi processing per se.

   Under these circumstances the software manufacturers should treat a socket like a socket no matter how many "scaled out" engines that socket contains.

   After all, if Intel produced a 5Ghz CPU with 20 execution pipelines and a sub processor to handle the thread execution in two blocks, but presented it to the OS as one CPU; what, functionally, would be the difference between that and Core? Except for the # of CPUs reported to the OS?

   So to reiterate, Core is about increasing power per CPU. That's why they call it cores per CPU and not CPUs per package.

   Software manufacturers should step up to the plate and accept that. After all, if Intel suddenly produced a single core CPU that was three times as fast as an existing single core product; what would IBM charge for it? Answer, One CPU. Core should be seen in the light of "smart execution engines" inside a CPU and not as an opportunity to charge more for additional CPU power.

   After all, software manufacturers have never been able to charge for additional processor speed.

   Also a point to note is that for single threaded execution, Core CPUs running at slower speeds than a traditional CPU, are always slower than the traditional CPU. In fact multi processing always carries a cost, so the software manufacturers are penalising the customers twice.

   First the CPU manufacturers sell them a CPU which requires multi processing (ergo slower on each individual transaction), to gain more speed; then the software manufacturers charge them more for the fact that the CPU “can” multi process even though the users don’t want it, they just want more speed.

   End result? Unhappy customer unless you take the Microsoft approach.

1. 4  Steve Rudland  | 7/26/2006 7:30:53 AM

   I wonder how long per processor or per processor value unit (pvu?) can survive anyway. IBM seemed to be quietly encouraging users aways from this model and onto the simpler CEO and Express models last year when they hiked Notes/Domino standard pricing by 26% but left the per user models pretty much alone.

   I was discussing this with a customer yesterday and they were far more interested in getting some flexibility into the user pricing. As a highly cyclical organisation they will typically have less than 200 users at the beginning of their business cycle (Year 1) and will scale up to over 900 users at the end of the cycle (Year 4 or 5). They will then drop right back to Year 1 again. They are keen to understand when the much discussed "software as a utility" will become a reality.

1. 5  Pete McPhedran http://www.corefusion.com | 7/26/2006 7:42:35 AM

   @3 - right on, except for one small caveat. Microsoft will be using this or a very similar approach soon enough. It was discussed at WPC06 recently and they just haven't made their announcement yet.

   Your points are excellent, I agree that the software manufacturers need to relook at their sales model in regard to CPU's. Specifically what value does Lotus or MS, or anyone else for that matter provide with the additional costs of the license? None.

   They do not allow for assignment of code to a specific CPU or core. i.e. run the HTTP thread on core 1 and if it fails, run it on core 3. There is no value add, yet they charge more.

   IBM also needs to add Service Provider licensing, but that is a whole other discussion...

   --Pete

1. 6  Curt Carlson  | 7/26/2006 7:42:48 AM

   If the goal is to get more money from less customers, this will do it. Our Customers want simple licensing models. Per Server is simple.

1. 7  Mark  | 7/26/2006 9:23:03 AM

   What would the model be if you ran 4 domino partitions on a dual processor (processors are both dual core) server?

   Would it be?

   DP * (P * PCU) = L

   DP = Domino Partitions

   P = processors

   PCU = processor core units

   L = Number of License

1. 8  Flemming Riis  | 7/26/2006 9:36:06 AM

   -They do not allow for assignment of code to a specific CPU or core. i.e. run the HTTP thread on core 1 and if it fails, run it on core 3. There is no value add, yet they charge more.

   You can set affinity yourself if the applications dont use multithreading but it will be a pain with restarts extra spaws , but software vendors just need get in gear and take advantage of it

1. 9  Charles Robinson http://cubert-codepoet.blogspot.com | 7/26/2006 10:57:29 AM

   At first I was outraged, but after reviewing the matrix it's apparent that pretty much everyone running Domino on Windows or Linux is going to end up at the same place they are today. It seems the people most affected are those running Domino as a shared workload on higher-end equipment.

   @6 - Unless you're on Express licensing you're not paying per server now. It's per CPU in the box, whether or not it's associated with Domino.

   @7 - Domino partitions are not licensed individually so that's not part of the calculation. I don't see anything on the PA page that describes how Processor Value Units (PVUs) correlate to licenses. In your example of a dual CPU dual core Xeon, that would be 200 PVU, the same as it would be for a dual CPU single core Xeon.

   The issue is if you run VMWare on this box you would still be at 200 PVU, regardless of how many instances or virtual CPU's are assigned to Domino. Domino is licensed per physical CPU, not virtual, which is the point Ed is asking for clarification on.

   I would like to see Domino only licensed per virtual CPU, up to the number of physical CPU's. So if you run 6 VMWare instances with Domino on a quad CPU box, you'd still only pay for a maximum of 4 CPU licenses... or 400 PVU's if they are Xeon CPU's.

1. 10  Ken Yee http://www.keysolutions.com/blogs/kenyee.nsf | 7/26/2006 10:58:41 AM

   What irks customers is if you change their former pricing when they don't change anything :-P

   The current Domino license treats an x86 dual core chip as a single processor and I've let customers know about this "feature" when they spec out hardware.

   As for long term, I'm not sure if processor value unit licensing will work that well...smells like that odd IBM mainframe/minicomputer model that works well if there's no competition, but w/ MS doing high-volume low-cost licensing, that doesn't work that well...

1. 11  NeilT  | 7/26/2006 12:10:24 PM

   @9 Yes, I agree, looking at the table the cost for Core Duo, Xeon or AMD X2 chips is flat as of today. However as both manufacturers scale out with 4 and then eventually 8 cores this model (once established), will begin to bite.

1. 12  Charles Robinson http://cubert-codepoet.blogspot.com | 7/26/2006 1:05:59 PM

   @10 - I don't have a reference handy, but I believe some of MS's products are licensed per virtual CPU. If you virtualize 10 servers onto a dual CPU box, you have to buy 10 licenses.

   @11 - That occurred to me, but I think we are a long way from higher core density becoming mainstream. As you said previously, more cores doesn't necessarily mean higher performance. Right now it's the hot new thing for the people who want to be leading edge, but when the business benefit doesn't follow the bean counters will cool things down.

1. 13  Paul Robichaux http://www.robichaux.net/blog | 7/26/2006 3:22:07 PM

   @12: I don't pretend to be an expert on MS licensing, which (like IBM's) is so complicated that you have to get a licensing specialist at the vendor involved to make even simple purchase decisions. I don't think there are any remaining products that require virtual CPU licenses, though.

   As for core density not becoming mainstream: I beg to differ. AMD is pushing hard to get multicore CPUs out there because they have a big performance/watt advantage relative to Intel's CPUs for the nonce. HP's data say that running additional cores equates to about 85% the performance of a physical CPU for Exchange, which makes adding cores a cheap way to scale up.

   For Exchange 2003, the sweet spot is 8 CPUs for most server roles. 4 dual-core CPUs will do nearly as well as 8 physical CPUs for heavy workloads, with substantially less power and heat (and less money, too!) I don't know where the sweet spot is for Domino, OCS, or any other applications, but I'm sure each app has one, and smart vendors will find out where they are and use that in their sizing guidance.

1. 14  Richard Schwartz http://www.rhs.com/poweroftheschwartz | 7/26/2006 6:39:58 PM

   Well, it's a start. But until it takes into account the actual core speed (or even just age) as well as whether or not you are using the entire CPU capacity or sharing it with other tasks on the same machine, it's still an inherently flawed model. And if does take those things into account, it will undoubtedly be too complex for pointy-hair types to understand. So why not give in to the inevitable and price based on use rather than (alleged) capability?

1. 15  Charles Robinson http://cubert-codepoet.blogspot.com | 7/27/2006 8:52:31 AM

   @12 - My MS rep told me that Win2k3 required a server license for each VMWare instance it ran on, regardless of the physical CPU's in the box. If you have contradictory information I'd love to hear it, but I would appreciate it if you could e-mail me so we don't clog Ed's blog with our unrelated discussion.

   Since the configuration you suggest is brand new nobody has it yet, so it only helps those people who are buying new. For everyone else getting to a quad-CPU dual-core environment is an expensive proposition. Hardware isn't free so using the same equipment across multiple software release cycles extends my value. That makes me and the bean counters happy.

1. 16  Alan Bell http://www.astoryforbedtime.com | 7/27/2006 4:32:36 PM

   how about making servers free, but raising the price of the client to a point that it would be revenue neutral, or at a modest net revenue increase. This would allow really really small installations to be financially viable, and you get more money out of the big servers supporting truck loads of clients. The benefit for the big guys is that they can relax on the server consolidation and the race for the biggest fastest server, and go for a more distributed networking architecture, small offices and client teams and individuals could have Domino servers running on laptops if they want. If you are building a big multiprocessor server you can then do so, without trying hard to minimise the core count and paying extra for speed.

1. 17  Ed Brill http://www.edbrill.com | 7/27/2006 4:38:31 PM

   @16 that's exactly the model for Express, and so it should allow "small installations to be financially viable". Right?

1. 18  Arthur Fontaine  | 7/27/2006 8:09:16 PM

   I didn't do the value unit pricing -- that's a Software Group-wide thing. But I was in the middle (with Ed and others) of most of the Notes/Domino packaging decisions over the past decade. The problem is that it always boils down to per-user vs. per-CPU pricing.

   Many customers like per-user because it's simple math -- count the users and write a check. It's especially appealing (from the KISS perspective) for SMBs because it avoids big start-up humps. (Domino Express offerings here.)

   Other customers like per-CPU because it puts them in control. The number of users per server is a dial that customers can use to get the best price per user. This is especially appealing to larger organizations that have a lot of users to support, as well as appropriate hardware resources for the task. (Utility Server offerings here.)

   But either one is hard for someone -- SMB's don't want to buy servers at thousands per CPU when they're supporting a handful of users; and enterprises don't like per-user because the costs are variable to infinity. So most of the products in our portfolio (certainly the Notes/Domino ones I have worked on) give you both options. Unfortunately it's still sometimes both options -- you buy both by the user and by the CPU. This is something of an artifact of how client/server software was sold in Notes' early days. Outlook/Exchange doesn't have this problem, but since most users pay for Outlook twice (once with Exchange and once with Office), that's easier for Microsoft to do.

   The art in designing an offering with multiple modes is finding where the modes naturally work themselves out, so it becomes a clear financial decision which mode to use. I actually like value units because they improve the granularity of how you sell server software, which can benefit both the customer and the vendor. Still, it only addresses the server part, leaving the basic difficulty intact.

   I will admit that no one in the industry seems to have gotten it completely right yet. Happy to hear all the ideas posted here. Say, do I hear a vote for Software Assurance-style annuity payments? I didn't think so. ;->

1. 19  Alan Bell http://www.astoryforbedtime.com | 7/28/2006 5:26:02 AM

   @17, I guess it is really. I had not realised that you could deploy as many servers as you like under Collaboration Express. I thought it was per user pricing for a single server organisation. So to expand my suggestion, how about removing the restrictions on Express (like clustering and the 1000 user cap) and get everyone onto Express pricing and make lots more noise about free Domino servers. I do kind of see the point about large organisations not liking variable costs, but with Domino it isn't hard to count the users each year and renew at that level, just don't force them to write a cheque each time they hire someone! My perception is that mega corporations never pay list price for anything though.

1. 20  Charles Robinson http://cubert-codepoet.blogspot.com | 7/28/2006 8:54:49 AM

   @19 - You hit the nail on the head. The Express licensing pricing is what I feel is appropriate for my 200 user organization; however, it comes with a list of restrictions that render Domino useless to us. It's like gelding a prize stallion.

1. 21  Mike Harris  | 7/28/2006 10:26:21 PM

   @19 and @20 - Express has been useful with many micro-customers but still falls short with many small SMB clients. Why is Directory Assistance an "Enterprise" feature, when cascading address books haven't been supported for years.

   A client can't use clustering with Express, but can have backup servers that replicate every 5 minutes. Most clients that use clustering use it for failover, not load balancing, so we make an easy job harder just for the sake of a licensing restriction.

   With PA Express only available to new clients, there's no transition for long term supporters stuck on PA model. e.g. One client with 200 users with 11 Enterprise servers due to geography and poor comms links. That ends up in major per user costs for this company.... Ripe for the Microsoft picking, at a time when when new Lotus customers can get PA Express with Trade-Up pricing from free-mail systems. I think we should spend a little time satisfying/protecting our existing customers as well as trying to win new user-base.

   This whole exercise reeks of people with too much time on their hands who should be out there actually marketing/selling Lotus to an ever increasing population who think MS is the answer... <End of Rant>

1. 22  Henning Heinz  | 7/29/2006 12:07:44 PM

   If we talk about Express, afaik it does not allow anonymous http access. So no web server or intranet service as long as not all users have a collaboration (express) license.

   I think the most attractive bundle for smaller companies still seems to be the CTO thing (if they have the minimum seat requirements).

   The current trend towards multi core technologies does not only have performance advantages but it is the current solution of AMD and Intel for lowering the enormous heat anticipation and power consumption of the bigger single core units.

1. 23  Richard Schwartz http://www.rhs.com/poweroftheschwartz | 7/30/2006 4:03:52 PM

   @22 You can do dual-licensing for a single organization, so you would purchase collaboration express per-uesr and utility express per CPU. Both licenses can apply to the same physical server, so you can have your anonymous http access on the same server that offers authenticated services to the collaboration users.

1. 24  Karsten W. Rohrbach http://photo.lytic.eu/ | 7/31/2006 5:53:07 AM

   [irony]

   So, how about per PSU Watts licensing?

   [/irony]

   And:

   "We're not announcing that today, but we're definitely headed in that direction."

   Sentences like this are bull*.*

   Either you announce it, or you don't.

   Those are the things that make customers and admins unhappy. Be deterministic, deliver what you say. Keep quiet when unsure. Everything else decreases credibility and affects your customer relations and sales figures directly.

1. 25  Tom  | 7/31/2006 6:37:58 PM

   IBM licensing is confusing enough,<sarcasm> hmmm.....lets make it more confusing, ya that will attract more business partners.</sarcasm> For example, Websphere, try and figure out what products you need for a specific purpose. Forget the installation of 5 products just to get Bowstreet Portlet Factory going. This has to be some MBA or PHD fresh out of college's idiot idea. Good news, this hair brained idea will help Microsoft as well as open source solutions.

1. 26  Henning Heinz  | 8/1/2006 8:13:40 AM

   Tom,

   Microsoft has an advantage as they sell boxed software and have some OEM preinstalled. That make it simple in some way but when it comes to enterprise licensing, with its armada of 500 or so products, it does not get any better.

   Open Source is another story but it is sad that the idea of Open Source is just reduced to free as in beer nowadays.