Hi,
> > Once you get above 8 they recommend something other than ethernet to
> > connect nodes (SCI interconnects etc.)
>
> What about gigabit ethernet w/ jumbo frames? Still not suggested?
I have 8 data nodes with gbit ethernet, and I have jumbo frames running. I
found it helped a little for large result sets, but allot of the packets are
still < 1500 anyways. I couldnt notice any real difference, although i did
not extensivly test it.
It seems like to get around some of the limitations it would be
> suggested to look at table designs and possibly design FOR the
> limitations - i.e. split up tables smaller so there's no 8k rows, not
> build tables with extensive indexes, etc...
I audited our table structure to ensure no wasted space (i didnt have any 8k
rows though), as the cost of having an excessivly large row in memory was
obvious. Especially with varchars being stored as chars (in 5.0) I had a
rapidly growing table so we developed an application to cycle out data into
another system for archiving as data over a couple of months old was
actually not needed on a regular basis.
Deploying on cluster isnt as easy as dumping data in and away you go. Fair
amounts of thought need to go into what your doing and how its setup.
Adam