Slashdot Mirror
OnLive Latency Tested
The Digital Foundry blog has done an analysis of recently launched cloud gaming service OnLive, measuring latency across several different games. Quoting:
"In a best-case scenario, we counted 10 frames delay between button and response on-screen, giving a 150ms latency once the display's contribution to the measurement was removed. Unreal Tournament III worked pretty well in sustaining that response during gameplay. However, other tests were not so consistent, with DiRT 2 weighing in at 167ms-200ms while Assassin's Creed II operated at a wide range of between 150ms-216ms. ... OnLive says that the system works within 1000 miles of its datacenters on any broadband connection and recommends 5mbps or better. We gave OnLive the best possible ISP service we could find: Verizon FiOS, offering a direct fiber optic connection to the home. Latency was also reduced still further simply due to the masses of bandwidth FiOS offers compared to bog standard ADSL: in our case, 25mbps."
The statement is silly because latency isn't directly related to bandwidth. Switches, bridges, repeaters, modems, routers and other such devices all add latency. If FiOS reduces the number of these in the chain, the latency will be reduced. I'm not saying it necessarily does - just that it could provide better latency without having more bandwidth because of other factors.
Yep, you're absolutely right in that bandwidth and latency aren't the same. However, when used by TCP in latency sensitive environments, common asymmetric connections can quickly saturate their available upstream bandwidth. This means that they're not able to ACK incoming packets, effectively increasing their link latency and reducing its throughput. So, in reality, total throughput is a combination of link latency and the ability to quickly respond to the protocol stream to keep the bits flowing. This is why QoS for TCP is so important on heavily utilised asymmetric connections.