This type of question is really too vague to answer without knowing what the data is and exactly what you want to do with it.
How much of the 20k of data items and 100k related objects do you actually have to scan/process over? If the 20k item is an opaque block like a photograph, for example, you probably aren't going to do much analysis of the block often -- you might pull the header or citation data and other related information down into 1k or less of useful data per item. On the other hand, if either the 20k or 100k are long form plain text, you might be able to distill it down to keywords to focus the data for efficiency.
When you say you want to do "multiple views on data, mostly in aggregate form", do you mean like OLAP cubes and other common reporting/analysis metrics, or more like searchy stuff where you might be interested in a needle in a haystack?
How do your users expect to interact with the data? If you were to sequential scan the entire uncompressed data set using Storagereview's current fastest disk, the Seagate 15k.5 300G, it would take over 21 disk DAYS ((min+max)/2 transfer rate), and this is contrived, and unlikely to be attainable in a real world scenario. It would take about 700 disks to hold it all, which would wall-time down to about 45 minutes. So naive walking the data for search purposes is out. It would have to be indexed.
Indexing is a tricky problem. In small databases they're usually so small and help so much that nobody pays any attention to the resources required to maintain them. With this dataset that's unlikely to be the case: you may effectively need several times the size of the data in indexing alone, which is going to have drastic implications on the computing resources required and the data storage format needed.
700 of those Seagate disks at roughly $200 street price is $140k. That's before simple redundancy, systems, cases, racks, power, logical redundancy for performance or extra space for derived data. Making a system to manipulate this much data is easily $500k, and paying a commercial vendor to provide you systems, hardware, and possibly software to do the job could be an order of magnitude more. For that kind of money I'd spend a few thousand hiring some people you trust to give you a reasonable assessment of the task at hand. It sounds like fun, though. Good luck. :)