96 active disks plus 4 spares configured in RAID1 gives 48TB of total storage.  So this basically gives 32TB of user space and 16TB of spool/temp space per rack, right?

With only 4 spares in a full rack, this makes it clear the spares are somehow sharable across the blades.  How is the disk actually configured?  I doubt that old method of duplicating data to a backup SPU was retained, so the most obvious configuration it is just set up as 48 mirrored pairs of disks.  I don't think that is actually how it works though.  Based off how Netezza is actually selling the system  with 3, 6, or 12 blade configurations, my guess is that the disks are configured as 4 sets of 24 disks in a RAID10 config with the spare only usable by one set.  This means the disk system is somehow acting sort of like a mini SAN.  I think this is the most likely config because it would allow another blade to access the disks of a failed blade thus preventing the system from completely failing in the event of a lost blade.  This method also greatly mitigates the problem of uneven data distribution between SPUs if all of the SPUs on a single blade share the same filesystem.

Ed said something about 'more than 1000 of these snippet processors' in his presentation.  What does that mean?  According to the datasheet the largest TwinFin system is 10 racks which is "only" 960 CPU cores + 960 FPGA cores.  Where does the 'more than 1000' come from?  Does the system run more than one snippet processor per CPU/FPGA core?

Can additional blades be added to a system on the fly or does Netezza still follow the model of just installing a second system and moving all the data to the new system?  That works perfectly well and really isn't a problem, but some people seem to think you have to be able to increase the capacity without any downtime.

Does losing a disk have any effect on queries which are currently executing?  I assume not since I believe the disks are now RAID10 disks instead of the old method of the data being copied to a backup SPU, but I just wanted to confirm.

What version of NPS will run on the TwinFin?  Is it Netezza 5.0, 4.6, 4.7, or what?

Other than greatly improved performance, scalability, and higher reliability, does the TwinFin have any other interesting features not available on the 10000 series?  I suspect the only answer I'll get for now will be "yes" and that I'll have to wait a bit to get more details, but I had to ask eh?