Data Warehouse Infrastructure

Question

I have a requirement to build a Data Warehouse and Analytics / Reporting capability with the following requirements...
Maximum of 1TB for Production Data + DR + Test/Dev Env.
SSIS (up to 25 sources), SSAS (cubes, 5 concurrent users) and SSRS (2 concurrent users, max 500 reports).
I needs a Production, DR and Test/Dev Environment 
I have been told that I will require 12 servers each having 4 cores and 12GB of storage (4 for Prod, 4 DR and 4 Test/Dev).
To give you an idea of load we plan to have 1 full time ETL developer, 5 Data Analysts, 2 Reporting Analysts. We are quite a small business and don't have a particularly large
amount of data. 
The model has SQL Server, SSIS, SSAS, SSRS on different servers across each Environment. 
Any idea if this is overkill? I also have an estimate of 110 days for Setting up the Servers, Installing the SQL Server software and general Infrastructure design activity.

Answer 1

Agree. Overkill. Big overkill.
I would recommend production/DR/Dev each have 2 servers. I'd put SSAS, SSRS and SSIS one one and the DB on the other.
In production, SSAS/SSRS will be active during the daytime; SSIS will likely be active off hours. So putting all that on one box should be fine for sharing the load. The DB on a second box would be good since it will likely be busy during the daytime
and night time. Four processors may be heavy depending on the types of queries and usage patterns. I suspect you can get by with 2 processor servers, but would recommend buying the 4 processor boxes for dev and production, get them configured and run
some performance baselines before putting in the DR environment. Then, if you find the CPUs idling, you can always cut the DR environment to 2 processor boxes. Not sure it's worth the minor cost savings to save 2 processors on 2 boxes with that effort, but
if you're looking to cut corners, you may find that a 2 processor per server DR environment is within your performance comfort zone.
For the dev environment, one box may well handle it all, but I'd go for 2. On average, a Dev environment isn't all that busy, but when you need the horsepower, you need it. And since it's Development AND Test, you help yourself by having realistic production
level performance on what you're testing. Four processors is fine, but max it out on memory.
As for hard drives, be careful about configuration. You need the space on your DW server and maybe for the SSAS server depending on how the cubes are built (ROLAP/MOLAP). When you speak about amounts of data, be careful since you'll want a lot of indexes,
and that can double the DB size for a DW. Your DW will also run faster if you have different filegroups for data/indexes/temp DB, but only if those different filegroups are on different physical media that work well in parallel. You can always get fancier
with more filegroups to have different ones for staging tables, for segregating fact & dimension tables etc. But for this size DB, that's overkill as well.
Mainly, I'd look at spending hardware $s on memory for the servers, but get less of them.
Now... two questions...
1) Can you clarify the disk space needs? How much total data space in one environment, without indexes? Based on that, add the same for indexes, add half as much (?) for TempDB and you have the core disk needs. Depending on how much it is,
you can decide on RAID, filegroup configuration, etc. And if the disk space with indexes is small enough that it all fits in memory, then disk and filegroup configuration becomes inconsequential except for ETL loads.
2) The 25 sources... can you clarify that? 25 source systems? Total of 25 source applications? Total of 25 tables? Curious, because I'm wondering about how long you'd keep 1 full time ETL developer busy.