I've planned to have secondaries on some of the larger locations (thousands of users), but after some reading and thinking I changed my mind.
What do you think about: We have a CAS and two primaries (200k clients), all located in Switzerland, same data center. There are remote locations (>200, not everyone gets a server) all over the world with some bigger datacenters, usually 1 or 2 per continent. Using a secondaries allows using fan out distribution and reduces the client traffic, but I figured the client traffic isn't a real issue. We have MPLS almost every, and mostly clients have to go through the last (and slow) mile at many locations already (CM07), hence the bottleneck is there either way. Having policy, inventory and SUP scanning to another continent increases latency, for sure, but isn't a bandwidth issue, it just takes like a second longer. On the other hand, as mentioned here also, it adds another tier to the hierarchy, adds complexity and it is a single point of failure for all clients using the proxy-mp or the sites using fan out. That is a rather huge disadvantage. What else to do? Connect the DPs directly to the primaries and use push, with the result that one slow box can basically kill software dist. for everyone. Or use pull DPs, which takes the load from the primaries, but requires to have bandwidth control on each DP. Both cause more traffic over the WAN, for every DP once. But I still think this is the better of the bad solutions. Yes, slower, but more reliable at the end. Opinions? -R
