Hello Ivan, On Mon, Feb 13, 2006 at 02:01:40PM +0000, Ivan A. Beveridge wrote:
> Hrmph - that was useless. Any reason why I'd not be seeing anything > useful in the trace? I thought it only looked like that if the binary > was stripped :( I compiled with '-g': Fully agree with you. I've just got a look to the gcc's manpage; taking an hint read there, try removing the '-O2' optimization flag, recompile and see whether results are somewhat better. Let me know ! "GCC allows you to use -g with -O. The shortcuts taken by optimized code may occasionally produce surprising results: some variables you declared may not exist at all; flow of control may briefly move where you did not expect it; some statements may not be executed because they compute constant results or their values were already at hand; some statements may execute in different places because they were moved out of loops." > Is there any way to see the default settings? (CONFIG-KEYS says that > default size depends on the OS). In "INTERNALS" file you mention about > "average_traffic = pps in network segment" - do you mean pps of sflow > data (in my case), or pps of traffic that will be sampled? Yes, computed values are returned commandline by the very first few output lines. To not get overflowed by all informations (caused by the arrival of sFlow packets), i suggest to check for those values attaching temporarily the collector to a "mute" interface. In your case, that PPS will mean the number of sFlow samples (not the total traffic). I've did some rough calculations: 30Gbps = 3.750GBps Assuming an average packet size of 700 bytes and a 2048 sampling rate: ( 3.750GBps / 700 ) / 2048 = ~2616 Then let's multiply for the sizeof(pkt_data) value. The 0.10.0 sees the growth of this structure, i've just updated the documentation. 70 bytes is a meaningful size (tolerating options: IPv6, 64bit counters, etc.). pipe size = 2616 * 70B = ~184Kb To let things work better it's strongly adviceable to batch collected data. You can do it by adjusting the buffer size. I have suggested you a buffer size of 10Kb. Going on with such small ratio (10Kb buffer size - 184Kb pipe size) is not that adviceable. Giving room to some 1000 buffers is usually prudent. Thus, 10Kb buffer size, 10Mb pipe size. Give it a try. > One trivial comment, with regards to text docs with the source - would > you make sure that line-length is < 80 chars, or it wraps horribly on > most people's terminals. :-) you are damned right :-) I will fix this. Cheers, Paolo
