I just ran a test of icecast to see how much cpu it uses. I set 500 instances of wget downloading from the mount point over loopback, and the cpu usage of the server was still barely reaching 0.3% on a 1.4G machine. The total bandwidth was 15Mbit/s. Each stream was 64 kbit/s. (I think some of the downloads weren't running fully - 15M/64k=240). I haven't tested over ethernet yet. What I would like to know is whether this level of cpu use is expected (seems very low), and generally what kind of spec machine we would need to serve up to maybe 100 simultaneous on-demand or real time streams?
On Oct 11, 2004, at 11:52 PM, Andy Baxter wrote:> What I would like to know is whether this level of cpu use is expected > (seems very low), and generally what kind of spec machine we would > need to > serve up to maybe 100 simultaneous on-demand or real time streams?Questions like this are inane and pointless. The answer is almost always the following: Buy the biggest CPU and as much memory as you can afford or wish to spend. CPU power is rarely the limiting factor these days. You kind of answered your own question anyways... I mean, you say you had 500 streams being pulled and the box wasn't taxed... and now you're asking about 100... am I missing something here? /dale
On Tue, 12 Oct 2004 00:20:55 -0400, Dale Ghent wrote:> On Oct 11, 2004, at 11:52 PM, Andy Baxter wrote: > >> What I would like to know is whether this level of cpu use is expected >> (seems very low), and generally what kind of spec machine we would >> need to >> serve up to maybe 100 simultaneous on-demand or real time streams? > > Questions like this are inane and pointless. > > The answer is almost always the following: Buy the biggest CPU and as > much memory as you can afford or wish to spend. CPU power is rarely the > limiting factor these days. > > You kind of answered your own question anyways... I mean, you say you > had 500 streams being pulled and the box wasn't taxed... and now you're > asking about 100... am I missing something here? >I thought there might be something wrong with the way I had done the test, because the CPU was way below what I'd expected. But I just worked it out roughly from the total bandwidth, and it does make sense.