3 ms·
@puzzle: while you're not wrong, do note that B4 is not (and is not designed to be) a low-latency, low-jitter network. It's designed for massive bandwidth for i
by packetslave 10y ago
@puzzle: while you're not wrong, do note that B4 is not (and is not designed to be) a low-latency, low-jitter network. It's designed for massive bandwidth for inter-datacenter data transfer.
- walrus01 10y agorunning your own internal links to near saturation (such as a theoretical 100 Gbps DWDM or MPLS circuit between two google datacenters in two different states) is a very different thing than running a BGP edge connection to saturation, such as a theoretical 100 Gbps, short reach intra building crossconnect from a huge CDN such as Limelight to a content-sink ISP such as Charter/TWTC or Comcast.
- packetslave 10y agoVery much so. B4 can be run near 100% because of strict admission control and optimized routing to maximize the use of all paths. It's much harder to do that on peering links where the traffic is bursty and you don't have control over the end-to-end latency and jitter. SDN isn't a magic pill for this, but it can most definitely lead to better performance and higher utilization than Ye Olde BGP Traffic Engineering.