SAN = Storage Area Network
SAN storage includes multiple servers connected to the storage network through HBA (host bus adapter), usually FC (Fiber Channel). In the most situations, servers are not connected directly to the storage but through one or two Fiber Channel switches. SAN is usually used in environments like high availability server clusters, virtualization environments like VMware Virtual Infrastructure, etc.
Also, SAN is much more manageable than DAS (Direct Attached Storage - disks in the server) and resilient to the failures, offering higher performances. The reason is that DAS is limited by the physical space in servers, while SAN devices always offer very good expandability (it depends on the class of storage). Since the storage server is responsible for disk drives management instead of the servers, any problems with the drives, expansion procedures, RAID management, different kind of replications, etc. are managed by the storage itself. The performance of the fiber channel is much higher than any other, because today you can choose from 4 Gbps and 8 Gbps infrastructure speed limits.
If you want to know about typical SAN devices you can check them out on the IBM System Storage web site. DS family consists of many devices for all business levels (from SMB to Enteprise).
NAS = Network Attached Storage
NAS storage is very simmilar to the SAN with one architectural difference. Instead of using expensive fiber channel infrastructure, NAS uses existing network infrastructure (Ethernet). Although, many NAS devices offer many different options for connections (including fiber channel). Also, NAS supports many different network file systems protocols.
NAS is used from home environments consisting of low cost 1 or 2 drives devices, to enterprise environments with capacities measuring in petabytes.
Maybe the best representative of the NAS architecture is the product line of NetApp.
Cluster server
Consists of two or more servers that works "as one". Clusters are used in mission critical environments (financial transactions, for example) where the availability of the applications are the highest priority job. There are, in general, two types of server cluster.
One is active/passive cluster where one server is always working, serving applications and data to the users, while other servers are "waiting" for that server to go off because of the hardware/software/other failure or maintenace. At the moment the server goes off, the next server (passive one) takes the data and applications and runs them. That situation is called "failover" and it is done automaticaly by the operating system, or manually triggered by the administrator(s). After first server gets back online, it can take the applications and data back to it ("failback"), or it can stay put, as the passive standby server in cluster.
The other type of cluster is active/active where two or more servers are working together sharing the tasks between them dinamically. The users are connected to the application, and never know which of the servers in the cluster are serving them. Servers that are members of the active/active cluster are dynamically share tasks between them trying to balance the load between them ("load balancing").
What technology for 0 downtime?
In general, it depends on what type of applications will you run on your system, what size of the environment are you expecting, how much money are you ready to spend...
Zero downtime can be reached in more than one way, although you must know that in the real world there is no absolute 0 downtime solution.
Not to make any detailde suggestions, but today, except very expensive mainframe systems, there are some interresting and powerfull solutions like VMware Virtual Infrastructure solutions for the high availability environments. For that purpose, VMware combines their ESX server with VMware VMotion, VMware HA and VMware DRS extensions.
