When architecting a highly redundant SAN, it is usually best (in my experience) to use either TWO fully meshed SEPARATE fabrics, or semi-meshed fabric with a core switch layer and a fan out layer for host access. Using fully meshed fabrics is fine for a smaller environment, but after you try to connect more than eight switches together, you end up loosing half the fabric ports to inter-switch links. This is where the core switches come in. By using at least two links to a core switch from each fan out switch and using TWO separate semi-meshed fabrics, the inter-switch connection issue does not become a problem. It leaves many more ports available for hosts while still providing a high degree of availability with a minimum hop count. In a single mesh, when connecting heterogeneous servers (NT and Unix), zoning becomes an issue due to less paths available to a particular server through the mesh if the zone is not configured correctly.
Then there are, of course, the director class switches. These are highly redundant in themselves and may also be fine for those with a large budget. Using two, of course, will be best and they make great "core" switches as the "backbone" of the fabric.
Most of the switch vendor Web sites have pretty good white papers on SAN design. Mark Farley http://www.digitalguru.com/dgstore/product.asp?isbn=0072120509&ac_id=69 and Tom Clark http://www.digitalguru.com/dgstore/product.asp?isbn=0201615843&ac_id=69 both have in depth books on the subject and are available at techtarget's Learning Zone http://www.digitalguru.com/dgstore/ttssto.asp?ac_id=69. Brocade has some real good stuff on their site at http://www.brocade.com/SAN/sansolutions.html, as does Hitachi on SWAN (Storage Wide Area Network) at http://www.hds.com/san/index.html.
There are frequent FREE seminars on the subject given by the switch vendors. Just visit their sites for times and dates in your location.
This was first published in May 2001