SharePoint Capacity Planning Tool Released - behind the scenes
The SCCP SharePoint Models were just released via the SharePoint team blog about an hour ago. Let me tell you about some of the behind the scenes on this one. Steve Peshka was really the main SME (subject matter expert). I admit I did have some input into it, but if it doesn't give you the right answer it wasn't my fault. :) You don't know how hard it is to build a model for capacity planning on SharePoint unless you've spent a few years with it and realized that this plastic reeeaaaaalllly stretches. Trying to tie down the WSS and MOSS PMs to get specific metrics for specific workloads in ranges is next to impossible. There is some magic to the tool which I think you'll find the more you dig into it. Some elements are static, some jump in ranges, and some plain don't exist. Memory for example is currently in the impossible to model category, so whether you put 32 GB or 8GB in the tool the data won't change as an example. Another thing, when we built it I wanted it to serve the 1 million users and 100 TB category since hey this SharePoint plastic can do that if you know how to configure it... hence the problem. If people are relying on a simple tool given a decent model, but multiple limitations they might over trust it. When people take our capacity planning guidelines on TechNet and follow them blindly without validating their scenarios, I cringe.
A SharePoint performance team in a lab with some software that doesn't know how one customer to the next will *really* use the product other than what limits they've been given they simply have to come up with tables for reference on recommendations or really provide some tested limits that are reasonable for the majority of customers. It's a tough job. Even trying to test with automated scripts... how large should the home page be? How many sites, how many documents, how large for the lists, what is reasonable? What kind of transaction mix should we expect? These black box tests minus your custom navigation, minus your actual file sizes, and your actual footprint... there are some major dependencies there. You do need to understand some very key elements. When you look at the layers of options and choices in the tool, I hope you're impressed that we did do a lot of work here to allow you to get as close to reality as possible without knowing what you're doing.
Not just latency and file sizes, but payloads, client configuration, network configuration, there are thousands of variables that play into an environment and NO TOOL can simulate everything. Not even SCCP version 15 2025. You'll always have some human variables and some custom development variables that will make testing still only be able to fit specific footprints and test specific loads. So my word to the wise is to not discount the value of your friendly neighborhood systems integrator, or your experienced SharePoint MVP from that blog you subscribe to. He likely could tell you more about capacity planning with a half hour conversation than any tool and write up on a napkin. I believe that. That's why it was an uphill battle to make a tool that doesn't do proper disk configuration and can't *really* simulate your network for you (because there are too many variables that even you can't find out). I tell you this because a lot of people are afraid of what trust people will put in this tool.
I am a fan of these new models and what basic guidance and topology it can provide because if it spits out the high availability 4 server WSS farm (2 WFE/2 SQL) for most configurations and the 5 server MOSS configuration (2 WFE/Query, 1 Index, 2 SQL) most of the time, it will send the majority of people down the right path. Sure a single server may meet the needs of small or medium business not looking for high availability, but they likely don't need to run the tool to find out that the single server is good enough.
It may seem like I'm rambling, and I am, but I hope the message that it there are thousands of variables and we couldn't think of everything and we had to stay on a tight budget and release before O14. You'll understand there were some constraints and I have a normal job as well. Thanks Satish for your patience with me, and driving to get the answers you did get to make the model as good as it is. I hope to hear that given all these constraints it still works well for the Intranets, the Internets and helps in scenarios we haven't thought of.
I think Satish's patience almost ran out when Doron joined the team and took a look at where we were with our models. Here we are with unrestricted limits and no red text telling what the limitations were. Throwing in some numbers and out pops no prescriptive disk configuration. Something he spent a lot of time on with the SharePoint SQL storage paper. How could we release a tool where we couldn't even control how the disks were configured. Put in 50TB and the tool would have all sorts of crazy SQL recommendations. How much do you put on a single SQL node? Is there a right answer to that question? Thanks Doron and WSS and MOSS PM leads for helping us narrow down to what we do know best (a lot of what we know from our top customers and what we learn from MS IT (thanks Bill Baer (capacity planning), Mike Watson (see his storage thoughts), and Corey of Reporting Framework Fame)). We'll leave the 100TB and 500,000 user environments for you smart architects to figure out and tell us what works best. If you run the tool enough times in a row, you can guess what our tool would tell you. ;) Really essentially that's what it would do... I had a customer complaint just yesterday saying the tool wouldn't tell them how it worked for their 300,000 users and I had to tell them, just run it 3 times. It would simply add more WFEs and SQL boxes with the same simple disk configuration. We have to blame some of these limits on the constraints of the framework, but then again can we really tell you how to optimize your disks if we don't know how much data and if it's at rest in an archive or needs to be super fast as all pages loaded up into memory and in cache in an internet site? We take advantage of using a common UI and a common framework and inherit a TON of hardware standards motherboards, CPUs, and tested hardware configurations and goodness from SCCP, but we also inherit some warts and limitations.
So far, in the hands of the MVPs and the 1000+ beta testers/customers that we had look at the tool, the feedback hasn't been overwhelmingly negative, actually opposite. This plastic tool given some structure can provide a basic model for determining some reasonable performance levels, even if we can't tell you how as you add RAM how that impacts your farm performance in the tool. One element and variable I'd consider more than critical to any deployment.
Get the SCCP tool and the SharePoint Models and let me know what you think. If you want more on capacity planning, see my key SharePoint Capacity Planning info organized by product and by scenario.
I failed to mention something that Paul mentioned in his great evaluation post on the tool. The SharePoint Capacity planning tool (models) and SCCP will run just fine on a Windows XP or Windows Vista machine. It uses a little bit of CPU during it's processing, but really is a pretty small footprint on your desktop. It's a desktop tool that every consultant, IT Pro, Architect should have in his toolbox. Add it your arsenal ;)
Joel
