Hello
as I read on other threads, I am experiencing low speed in single tcp connections test.

I try with iperf through one RB1009 and one 1036, and another 2004 with rj01 (not the eth port)

Multimple tcp stream test, I can saturate easily 1G speed.
Single tcp test, the best case are around 500M

I tried both with fastpath (routing)
fasttrack
no change

I have read that the connection is tied to a single CPU, so the TILE are around 1000-1200MHz, I replaced them with a 2004 with 1.7Ghz ARM cpu, I expected better results, but they are the same.

how can I solve this issue?

For me, all these single TCP session throughput issues always boiled down to packet loss so far - either caused merely by low quality network or by a too coarse bandwidth shaping. So watch for this at first.

Hello, thank you for your reply.
In that part of the network we don't have any shaping or bandwidth restriction.
I am able to reproduce this issue with single tcp connection speed (as other in the forum).

Have you tried to sniff at the iperf server and iperf client themselves? I don't comment on CCR10xx as there indeed the concept of many relatively weak cores may affect single-stream throughput; I have in mind when testing with the CCR2xxx.

Hello
what do I have to sniff?

how can I solve this issue?

Try replacing hardware routers with CHR routers.

It is not possible. However both 1036 and 2004 have the very same issue.

However both 1036 and 2004 have the very same issue.

Yes, all hardware routers have same issue.
CHR do not have this issue: viewtopic.php?t=146665#p770846

what do I have to sniff?

The idea of sniffing is to see whether the throughput is limited by delays between packets or by losses and subsequent retransmissions. So you should sniff the test TCP session and then use Wireshark to look for anomalies.

Thank you. There are no limitation in this aspect.
the limitation appears to be on the MT hardware routers

@normis any news?

Is iperf showing dropped packets / retransmission? If not then it's limited by OS on one of the sides. Have you adjusted buffer sizes and window scaling on the sender / receiver?

No losses anywhere. If i replace the mt with a cisco a edgerouter for example, i dont have this limitation

Strange, I've managed about 960mbps through a single TCP connection on a CCR1009. What does tool / profile look like?

Try doing nat on the ccr, with fasttrack.

any updates on this issue?
this is a huge problem..

I replaced the CCR with a ARM cpu router (as the 2004) and the issue disappeared.

I can confirm that it is way better on arm device like 2216.
yet we detected weird problem if we connect from different interface of ccr and test single stream downloads we see 200-500 mbits even if there is 8 gigabit of traffic load but if we do not connect directly to that router for example we connect from a device that located behind that router and if there is a traffic load like 5-7 Gbits on 10G port, then, single stream speed reduces to 20-10 mbits ..
we tried every possible configuration fast path, fast track without any success.