On Thu, 2007-11-01 at 11:09 -0400, Sam Hartman wrote: > In many cases the performance of security protocols is not a huge > issue at all with modern hardware. Depending on the scope of this effort I see a bunch of things that might be worth modelling: - additional compute resources (cpu, memory, crypto hardware, battery energy, etc.) - additional round trips (that pesky speed of light thing) - increases in message sizes/reduced effective MTU - availability (if your security protocol depends on a KDC/OCSP server/CRL distribution point/other service, if it's not available, you're not available) The first is not a factor on typical client computers, but the same is not necessarily true for the mobile phone/pda class of widget or even for servers -- in the latter case customers say they want high enough utilization that the overhead of security protocols is going to be significant for server sizing. > It has not been my experience that it is important to a level where > metrics are requested or used. it's very common for customers to ask "how big a server/server farm do we need to support this expected workload?". The impact of security protocols on that workload can be significant. - Bill _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf