Yes, it is recreated which each server start. But I thought, that fail over servers allows me to fail over without interupting every connection. When I restart my database server, I have to stop and start the applications, that are using it, too, because the most applications needs some init stuff and does not really like it, when the server connection get lost. I may be too optimistic (we are not using clusters, so I have no experience), but a failover should be transparent to the apps (they should not even realize, that there happens something)

@jeff Modem: of course fixing the code would usually result in much more improvements. But going to Amazon, ordering 3 customer SSDs (hey, its only TempDB), putting it into the server and creating a RAID 10 would be maybe 1000 EUR / USD and increases the speed immediately. Fixing the code in a big application could cost several man-years of work - if you can change it (false for the most vendor apps and of course you need somebody who is able to do it)