Dramatic bandwith impact when chaining multiple LS1021ATSN devkits

cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Dramatic bandwith impact when chaining multiple LS1021ATSN devkits

573 Views
sebastien_merle
Contributor II

For our prototype we are required to daisy chain multiple LS1021ATSN routers.

We are measuring the bandwidth by downloading a very large file over HTTP either directly between the client and the server or passing through different configuration of routers (numbered from 1 to 3):

|Direct | 112,00 MB/s (stable)       |
|1      |  52,80 MB/s to  98,00 MB/s |
|2      | 112,00 MB/s (stable)       |
|3      | 112,00 MB/s (stable)       |
|1+2    |  22,60 MB/s to  54,00 MB/s |
|1+3    |  19,00 MB/s to  69,30 MB/s |
|2+3    |  23,80 MB/s to  73,2 MB/s  |
|1+2+3  |   2,89 MB/s to   2,95 MB/s |

First we can see that router number 1 is acting strangely. this is the same router we keep having issue with when updating the configuration. It seems that router 1 is defective in some ways.

Second the drop of bandwidth when chaining multiple routers seems quite exaggerated. Even considering the configuration that include router 1 that is probably defective, the configuration with only router 2 and 3 chained shows a drop of 35 to 80 percent of bandwidth.

And finally, the measures were done downloading a 10 GB files and the average bandwidth swings wildly from one measure to the other. This could be related to the TCP congestion algorithm and the packet drops, but it still looks like too much variations to me.

Is there a way to reduce this dramatic lost of bandwidth ?

Attached are the configuration of the routers when chained together.

Thank you for your support.

0 Kudos
1 Reply

464 Views
alexander_yakov
NXP Employee
NXP Employee

Please answer the following questions:

  1. What does the hardware setup look like (what are the endpoints, how are the cables connected, and what devices do the packets pass through)? A simple picture would help. The term "router" has been used, which suggests to me that some IP forwarding is taking place (iptables, etc). But only some L2 switch configurations were provided.
  2. Compared to a standard SJA1105 configuration, the provided XMLs have changes in the following areas: (a) increased l2-policing-table->maxlen to 2043: presumably to support VLAN-tagged packets and not drop them (b) configured VLAN ID 1, 0x26C 0x26D and 0x26E on the switches and made 0x26D as the default (c) enabled MAC filtering for 01:80:c2:00:00:14 (d) configured a mirror port. Which of these settings are necessary in order to have an adequate image of the problem, and which can be ignored?
  3. What type of synthetic traffic would be adequate to use to simulate the customer's use case? Would iperf3 in TCP mode do? Is the traffic being sent by the endpoints as VLAN-tagged or untagged? What about the traffic between the switches? If the traffic has more route segments than just endpoint1 -> sw1 -> sw2 -> sw3 -> endpoint2, detailing the frame format for those extra segments is important too.
  4. What is the MTU on the devices configured for?
0 Kudos