Brad: I've been curious about the cache hit/miss data also, as well as the overall cache architecture (havn't found any info), but I don't necessarily agree that the cache is "crucial to providing high bandwidth P2P under the cable modem network's strategy." The attachment to the backbone is more key to the speed. Also, depending on the architecture, if a read miss occurs, it may be such an insignificant amount of time relative to the next available time slot in the multiplex stream that it is imperceptible except under extremely heavy loading (but I have no facts to back up this theory). Also, the downstream data path is really point-to-multipoint (or one-to-many), so a capacity limit can be reached--in theory. I've seen several articles about the studies done regarding a slowdown due to capacity overload, and they all claim (because they have a vested interest in claiming so, and no one can prove them wrong) that it's not a problem providing this condition and that condition aren't violated. For the most part I consider the test conditions unrealistically simplistic (short bursts of light bandwidth,spaced at roughly 60 second intervals, times some number of active peak-demand users) and the claims sort of misleading because they only give data for 500 or fewer drops per node, which is the absolute minimum in the system models I've seen. I can't really refute their peak usage models because they are based on projections, but most use roughly 30% peak subscriber load, which seems a bit light to me if this type of service takes off the way we're all hoping, and content bandwidth increases the way everyone is projecting.
dh |
