Blog entries by Herco van Brug

we did some tests at a customer last week with FT. First ofcourse, it works like a charm. very impressive technology.

one of the things we wanted to know is the bandwidth needed to keep the CPUs in sync.

first we popped up 'calc' and did a 123456789!  CPU went up to 100%, bandwidth needed: only 15kB/s.. wow.. 

next we did a local disk to local disk copy. not very cpu intensive but the FT bandwith went up to 25kB/s. still very low.

next we did a couple of things at once. dir /s from a console, excel sheet operations, all kinds of different loads, all at once. now the bandwidth went up to 2500kB/s. ok, that's something else! somehow it seems to be very efficient with small cpu repetitive instruction sets but when there's a lot of context switching, it really needs its bandwidth. but that's not the end of it. after a couple of seconds, the bandwidth went down.. considerably.. at the end it only used 56kB/s again. so... wait a minute, it seems to be learning what it's doing and will only send differences over the network? that's very clever.

so in rare peekmoments FT really needs its bandwidth. but the average bandwidth needed is really very low. latency seems much more important than bandwidth. ofcourse, if you protect 20 VMs you have to take into account that they may all need a lot of bandwidth at a certain moment, but it's smart. it seems to be learning, making sure it doesn't resend everything the other end already knows, caching it somehow.

don't you just love technology you have no clue about how it works? i'm impressed ;)