I can't imagine fielding 2MP, HD quality video, or better without the use of Smart CODECs. H.264 yields about 50% over MJPEG. We have measured as much as another 80% savings with the Hikvision Smart CODEC for video with no motion. We really hope Hikvision offers the automatic frame rate elevation on video motion detection which will also offer even more dramatic savings as well (Axis just came out with this). Presently we are using the Milestone rule engine to elevate our frame rates from 10 fps to 30 fps on motion and that works pretty well but we would rather the camera provide this processing which will cut down on network bandwidth.
We are seeing streaming bit rates of 300 kbps to 600 kbps for HD video (50 out of 100 quality points) at 30 fps. This is remarkable since we are replacing analog video encoders with only 4 CIF resolution that are streaming at 1000 kpbs using MJPEG.
The pink elephant in the room though is the computational intensive decoding that must take place on the workstation clients. It is bringing our server-class HP workstation to it's knees. Our workstation has two processors with 14 cores each. With hyperthreading this gives us 56 threads of execution. It can't decode more than about 16 cameras live at 30 fps. Our customer expects up to 75 cameras per workstation spread across three monitors. We really hope the VMS manufactures address this limitation. Milestone has started down this path supporting Intel Quick Synch processors but unfortunately that processor line is limited to the lower end (single processor, four core). Leveraging the GPUs engines in AMD and NVidia graphics cards could really allow scaling of this problem while still using a more reasonably priced workstation.