Thanks Bruce. I had to do MSS mangling to avoid fragmentation of packets after ESP header addition.
With single streams I get about 1/3rd of what I get using multiple streams Thanks and regards Rajesh > On May 12, 2018, at 2:34 AM, Bruce A. Mah <b...@es.net> wrote: > > If memory serves me right, Kaliaperumal, Rajesh wrote: >> Hi Bruce, >> Single stream was pretty low and then I started trying multiple streams. >> I am expecting a throughput of the order of 70-100Mbps (my ISP DL being the >> limiting factor). > > Hmmm. I don't have any good explanation off-hand. The only two > thoughts I can think of right now are: 1) the IPsec tunnel and MSS > manipulation are unusual features of the path, and 2) that's a fairly > old version of iperf3. I'm not sure if either of those are a factor, > just saying they stand out to me. > > Note that using multiple streams isn't necessarily going to scale the > throughput linearly. > > Bruce. > >> On 5/9/18, 4:02 PM, "Bruce A. Mah" <b...@es.net> wrote: >> >> If memory serves me right, Kaliaperumal, Rajesh wrote: >>> Hi all, >>> >>> >>> >>> Iperf version – iperf 3.1.3 >>> >>> >>> >>> My set up: iperf server (GCE) <=> Gateway VM (GCE) <=> IPsec server >>> (GCE) <=> WAN <=> Laptop with IPSec Client >>> >>> >>> >>> *My test:* >>> >>> Server – /iperf3 -s / >>> >>> Client – /iperf3 -c <server ip> -P 15 -R -w 768k/ >>> >>> Due to IPsec header, my Gateway VM modifies the MSS size of the >>> SYN-SYN/ACK message so that the TCP segments are not fragmented after >>> IPSec ESP encryption. >>> >>> >>> >>> *My Observation:* >>> >>> When I perform a downlink test by using the “-R” option at the iperf >>> client side, my throughput has a lot of variation and quickly drops to 1 >>> or 2 Mbps. In other words, the iPerf server reduces the rate at which it >>> pumps data. There are no packet drops on other VMs in my setup. >>> >>> >>> >>> I want to understand how iperf server determines the rate at which it >>> has to pump data? >> >> With those parameters (and thank you for providing that information), >> the server basically just sends data as fast as it can on the 15 TCP >> connections you specified by the -P parameter. The speed at which it >> can send data is governed by the TCP congestion control and flow control >> algorithms. >> >> Of course those TCP connections will interact with each other, and any >> other network traffic along the path. If the system isn't behaving in a >> way that makes sense to you, might I suggest trying your experiment with >> just a single TCP connection? Also, what throughput were you expecting >> to get? >> >> Bruce. >> >> >> >> > > ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot _______________________________________________ Iperf-users mailing list Iperf-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/iperf-users
