Clouds-Colors_680x340Our friends over at Cloudharmony just published a new benchmark, this time on inter-cloud throughput. I think they are doing a great job at trying to provide some much needed transparency on IaaS performance. In their last benchmark (which ranked throughput to end users), Voxel was ranked as the fastest cloud in the United States (IBM and EC2 came in second and third). So, how did we fare this time around? I’m pleased to report that we set the speed-record for inter-cloud transfer — 183Mbit/sec between our cloud in NYC and Linodes cloud in Newark NJ, to be specific. While we’re of course thrilled by that fact, there are some points that should be made that help explain this result:

1) Latency matters! The single biggest factor that can help (or hurt) network throughput is latency. This is why the speed-record was set between two clouds that are across the Hudson river from each other. I can see downtown Newark from our office, and Linode is only a few milliseconds away from us, from a network perspective. You can’t sling packets faster than the speed of light, so it helps to be close. Voxel is the only cloud provider to offer service out of the United States, Europe, and Asia, so pick the location that makes the most sense for you and your users.

2) A lot of clouds rate-limit at 100Mbit/sec, or less As CloudHarmony pointed out, many clouds rate limit individual virtual machines to 100Mbit/sec, a ‘cloud speed limit’ of sorts. With VoxCLOUD, the limits are much higher (as they are at Linode too, apparently). At Voxel we like to joke that QoS stands for Quantity of Service — it’s in our DNA to massively overbuild our network. Maybe operating the VoxCAST CDN has something to do with that.

Next week, we will be making some network-related updates to VSE (the Red Hat based “Voxel Server Environment”). These updates and tweaks will bring further improvements to network performance and throughput, driven from optimization to the Linux stack. Next week also brings a new version of VoxCLOUD (1.10). More about both in a future post, stay tuned!