USG1100 - performance expectations for routing between vlans?

imaohw
imaohw Posts: 123  Ally Member
First Anniversary 10 Comments Friend Collector First Answer
edited April 2021 in Security

With a USG1100, what performance (in Mbits/sec) should I expect when testing using iperf3 (tcp) with the server on one vlan and the client on a different vlan?

The two vlans are associated with different Base Ports.

All of the UTM features of the USG1100 are turned off and there is no BWM.

Minimal other traffic on the USG1100 and CPU Usage < 2%.

«1

All Replies

  • PeterUK
    PeterUK Posts: 2,655  Guru Member
    First Anniversary 10 Comments Friend Collector First Answer

    I would think you get the full 1Gb (not including over head) if not by a single connection thread but more at the same time.

  • imaohw
    imaohw Posts: 123  Ally Member
    First Anniversary 10 Comments Friend Collector First Answer

    @PeterUK - I expected close to 1Gb with a single thread but come nowhere near that. I am seeing 200-400 Mbits/sec with one thread. With four or more threads I can sometimes get close to 850 Mbits/sec. But not above it.

    I asked the question because I was surprised it is so low with a single thread and wasn’t sure my expectations were reasonable.

  • PeterUK
    PeterUK Posts: 2,655  Guru Member
    First Anniversary 10 Comments Friend Collector First Answer

    How many cores does the USG1100 have ? you can check this in in diagnostics > CPU/memory status.

  • imaohw
    imaohw Posts: 123  Ally Member
    First Anniversary 10 Comments Friend Collector First Answer

    CPU0 - CPU5. So I guess 6 cores.

  • PeterUK
    PeterUK Posts: 2,655  Guru Member
    First Anniversary 10 Comments Friend Collector First Answer

    Maybe there is a issue with load balancing the thread across the cores its not like you got two VLAN's on the same Base Ports.

    If you have a intel NIC maybe try with “adaptive inter-frame spacing” enabled on the sending device.

    Or if you got switches to the USG1100 try without and devices to the ports of the USG1100.

    Is MS file sharing slow too?

  • imaohw
    imaohw Posts: 123  Ally Member
    First Anniversary 10 Comments Friend Collector First Answer

    @PeterUK - thanks for the suggestions. I started this experimenting when I was seeing lower than expected throughput between vlans. Both of the USG1100 ports are connected to ports on the same slot on an XGS3700 switch stack.

    If I put both test machines on the same Vlan (just going thru just the switch- not USG1100) I get 985 Mbits/sec.

    I can’t test with vlans and the devices connected directly to the ports on the USG1100 as the devices (server and pc) are not Vlan aware. I would need to test from Ethernet port to Ethernet port. To do that I need to configure those ports slightly differently. Once done I will post the results.

    In the meantime, let’s see what Zyxell says I should expect for throughput results.

  • Zyxel_Jerry
    Zyxel_Jerry Posts: 1,026  Zyxel Employee
    First Anniversary 10 Comments Friend Collector First Answer

    Hi @imaohw

    If the case is one session transmission(the same src IP/port and dst IP/port), it will keep running on a core.

    The device will use one core to do the packet forwarding and this is the current design mechanism.

    However, when running the test with multiple sessions, you can get higher transfer speed and you will be able to see the multiple core handling the transmission.

    When doing iperf test(using single session) cross different vlans, the throughput will not be as fast as what you’ve seen when just going through the switch.

    However, the throughput will be much higher when there’re multiple sessions.

    In our test, using iperf test cross different vlan can reach nearly 920Mbits/s. with multiple sessions.

  • imaohw
    imaohw Posts: 123  Ally Member
    First Anniversary 10 Comments Friend Collector First Answer
    edited February 2020

    @Zyxel_Jerry - I have a few follow up questions:

    • In my test, the source and destination ports are different. One Vlan is associated with one port the other Vlan is associated with a different port. Would the USG1100 to be using two cores?
    • What throughput in Mbits/second should I expect to see with a single session iperf3 test across two ports?
    • If I am using a LACP LAG between the USG1100 and a switch for a similar test as above, with two vlans, is the number of cores used based on the number of physical ports utilized for traffic in and out or is the LAG considered one “port” for deciding how many cores to use?
    • What is the maximum throughput in Mbits/sec of a single core on a USG1100?
    • The spec sheet for the USG1100 shows throughput of 6000 Mbps. How could that be achieved if a single core is only capable of approximately 300 Mbps and there are only 6 cores?
    • Can you describe the test that was used to produce the results in the USG1100 specs for throughput?
  • Zyxel_Vic
    Zyxel_Vic Posts: 281  Zyxel Employee
    First Anniversary 10 Comments Friend Collector First Answer

    Hi @imaohw

      In my test, the source and destination ports are different. One Vlan is associated with one port the other Vlan is associated with a different port. Would the USG1100 to be using two cores?

    Reply: No, the same IP/Port pair (a session) will be processed through 1 specific core

    ·         What throughput in Mbits/second should I expect to see with a single session iperf3 test across two ports?

    Reply: 300Mbps up can be expected

    Connecting to host 192.168.7.50, port 55555

    [ 4] local 192.168.8.50 port 3258 connected to 192.168.7.50 port 55555

    [ ID] Interval          Transfer    Bandwidth

    [ 4]  0.00-1.00  sec 43.2 MBytes  362 Mbits/sec

    [ 4]  1.00-2.00  sec 43.9 MBytes  368 Mbits/sec

    [ 4]  2.00-3.00  sec 43.5 MBytes  365 Mbits/sec

    [ 4]  3.00-4.00  sec 48.6 MBytes  408 Mbits/sec

    [ 4]  4.00-5.00  sec 45.1 MBytes  379 Mbits/sec

    [ 4]  5.00-6.00  sec 47.0 MBytes  395 Mbits/sec

    [ 4]  6.00-7.00  sec 45.6 MBytes  383 Mbits/sec

    [ 4]  7.00-8.00  sec 44.0 MBytes  368 Mbits/sec

    [ 4]  8.00-9.00  sec 46.1 MBytes  387 Mbits/sec

    [ 4]  9.00-10.00 sec 43.8 MBytes  367 Mbits/sec

    - - - - - - - - - - - - - - - - - - - - - - - - -

    [ ID] Interval          Transfer    Bandwidth

    [ 4]  0.00-10.00 sec  451 MBytes  378 Mbits/sec                 sender

    [ 4]  0.00-10.00 sec  451 MBytes  378 Mbits/sec                 receiver

     

    iperf Done.

     

    ·         If I am using a LACP LAG between the USG1100 and a switch for a similar test as above, with two vlans, is the number of cores used based on the number of physical ports utilized for traffic in and out or is the LAG considered one “port” for deciding how many cores to use?

    Reply: No, since there is still a session in certain scenario. The performance will be the same with when running without using LAG

    ·         What is the maximum throughput in Mbits/sec of a single core on a USG1100?

    Reply: Around 350~400 Mbps maximum

    ·         The spec sheet for the USG1100 shows throughput of 6000 Mbps. How could that be achieved if a single core is only capable of approximately 300 Mbps and there are only 6 cores?

    Reply: This is tested by using 3rd party testing hardware to generate full speed udp packets(with different sessions) in fully meshed scenario. For instance, if the incoming is 800Mbps, outgoing is 800Mbps, the accumulated backplane throughput will be 1.6Gbps on this port. To achieve the speed mentioned in the datasheet, a dedicated hardware testing tool is required.

    ·         Can you describe the test that was used to produce the results in the USG1100 specs for throughput?

    Reply: We tested the devices with a 3rd party test machine which runs RFC2544 to do the throughput test.

  • imaohw
    imaohw Posts: 123  Ally Member
    First Anniversary 10 Comments Friend Collector First Answer

    @Zyxel_Vic - thanks for all of the information.

    If I understand correctly:

    • Each “session” will be handled by a single core.
    • Each core is capable of 350-400Mbps throughput (tcp).
    • There are six cores in the USG1100.

    Realistically the maximum throughput I should expect from the USG1100 is around 2,000Mbps - 2,500Mbps (tcp)?

    I also have some Zywall 310s. I see they have six cores like the USG1100. Is the difference between the 310 and the 1100 that the 1100 has a slightly faster processor?

Security Highlight