Does the network sag from small loads?

Question

Computer networks

I do server stress testing with Apache Benchmark Tool and Apache jMeter. With 30 or more concurrent requests, the network periodically (approximately every 100-200 requests) sags for 3-4 seconds.
With all this, there is no load on the server, LA is zero, there is plenty of memory.
those. 200 requests go with a 50ms response, and then ~10 go with a 3000ms response, and then the situation repeats.
And this does not depend on the content in any way, because the tests were even on static.
Server config:
Leaseweb
Intel Xeon X3440/8GB/2x1TB (QC22)
QC X3440
8GB DDR3
1 x 100Mbps Full-Duplex
What could be the problem and how can I monitor weaknesses?

Answer 1

What did you get that it is the network that sags? Try to ping the server in parallel with the stress test (to the fullest). If under load the pings are stable, then everything is in order with the network. And Apache can have 100 reasons why it slows down.
Look at the output of top under load to begin with, how long it sits in the kernel, what processes are at the top, how the swap is used.

Answer 2

check ping (L3) and TCP level (L4)
if everything is OK, then dig on L7 (apache)

Answer 3

put some kind of server monitoring system, for example, munin, cacti, zabbix, etc. according to the graphs, you can see what resources are missing.
you didn't write anything about the software on the server. Do you have apache running there? Periodic failures can be related to MaxRequestsPerChild - for example ...

Answer 4

> Periodically (approximately every 100-200 request) the network sags for 3-4 seconds.
what you described is similar to the rebirth of the worker.
"MaxRequestsPerChild controls how frequently the server recycles processes by killing old ones and launching new ones."