A stretch cluster is one in which nodes are separated - possibly by several miles - mainly as a precaution against a complete data-centre outage. I know of a couple of examples where separations in the region of 20-30 miles (30-50 km) are either in use or planned, and other posters on Oracle-L have mentioned "several" or "a handful" of implementations worldwide.
Obviously, the main performance issue for a stretch cluster is the latency and bandwidth of the interconnect. The bandwidth isn't affected by distance, but the latency certainly is.
I'd be very interested to hear from anyone who has implemented a stretch cluster (in test or production) with as much detail as you are free to pass on, particularly:
- what is the distance between sites
- how many nodes at each site
- is the workload evenly distributed (active/active homogeneous), partitioned (active/active, heterogeneous) or uneven (active/passive)
- some indication of database size and transaction rates (in whatever units are meaningful to you
- Any performance issues?
Either email me (nigel at preferisco dot com) or comment below. I would like to publish the results but let me know if any part(s) of your information is too sensitive to be broadcast, even anonymously.
Thanks in advance...