On 2014-01-24 16:18, Ruben Laban wrote:
On 2014-01-24 15:56, Paolo Lucente wrote:
Interesting behaviour you are describing.
In-line for a comment about the high interface value:
On Thu, Jan 23, 2014 at 09:01:56PM +0100, Ruben Laban wrote:
As for the ports, I meant to say interfaces ;-) So I did a snmpwalk
against the switch and this told that those interface numbers 211
and 48 correspond to respectively trk2 (2nd configured trunk) and
port 48. So far, so good. The number 1073741823 doesn't show up at
all in the snmpwalk though, which is rather odd. Then again the
number is 0x3FFFFFFF which is probably something "special".
Indeed, that is a "special" indicating that there is no input/output
interface (depending which field the 0x3FFFFFFF is found). This is
typically the case if you ping the switch itself, for example.
Which seems like a bug in the switch then (wonder if it turns out to
be one of many...). As you can see, it's the reply traffic for the
packets with the proper interface IDs.
Based on some more testing: it appears that the switch offers very
little CPU cycles to the sFlow engine. With a sample rate of 1000 I
don't get dropped samples. With a sample rate of 100 and lowered
traffic throughput it also copes without dropping. But even with no
drops being reported, the numbers I see in sfacctd are still way too
low. I have no reason to blame sfacctd just yet, based on the other
stuff I've seen.
So far these switches (or at least the sFlow part of them) aren't
giving me much of a comforting feeling. Next week I'll try to reach
out to HP regarding these issues.
Well, after having been in contact with HP technical support, I'm not
sure if my confidence in these switches has increased much. Their
response pretty much was: "We're not seeing any technical issues with
your switches, so if the observed performance is not what you expect,
you should have bought more expensive switches".
Currently I'm running some more tests to figure out the "exact"
limitations.
Raw info so far (mainly just FYI):
57700 fps - 1730000 frames - 30seconds
1/50 : 33613 dropped / 34600 expected / 987 sent?
1/100 : 16029 dropped / 17300 expected / 1271 sent?
1/1000 : 889 dropped / 1730 expected / 841 sent?
1/1100 : 612 dropped / 1572 expected / 960 sent?
1/1200 : 0 dropped / 1441 expected / 1441 sent?
1/1500 : 0 dropped / 1153 expected / 1153 sent?
1/2000 : 0 dropped / 865 expected / 865 sent?
(The question marks are there because those number are calculated based
on the rest, not the actual amount of received samples)
Regards,
Ruben
_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists
