Quote:
Originally posted by stief
Are upstream stats reliable? This is an old problem, but just in case it's more than a GUI problem, here's a screenshot. (14 hrs uptime; uploads throttled to 101.25 KB/s on a 1 mbps cable connection) |
The upstream bandwidth currently only counts the actual TCP payload (ignoring automatic retransmissions in case of TCP errors detected), but not the TCP header or IP headers which are unpredictable due to buffering within the OS, and not even the PPP/LCP overhead on the internet link or the ATM encapsulation.
To get more accurate data, you need to estimate the overhead, or look into OS statistics (you won't get data below the IP level (i.e. PPP/LCP or ATM framing will not be visible on DSL/cable connections; on 56K modems, you can get the final bandwidth sent to the modem in the OS statistics, but even in that case your modem will apply its own protocol (data compression, error control, and framing bits).
The true bandwidth used is hard to determine (you can just take an estimate of about 5% of overhead for the underlying transport or link protocols).
Also the "spiky" bandwidth is typical in case of slow or far downloaders, that can't reply in sustained time. This spiky curve is smoothed when there are parallel transfers to distinct hosts.