• Categories
  • Recent
  • Tags
  • Popular
  • Users
  • Search
  • Register
  • Login
Netgate Discussion Forum
  • Categories
  • Recent
  • Tags
  • Popular
  • Users
  • Search
  • Register
  • Login

Is my CPU limiting the throughput?

Scheduled Pinned Locked Moved Hardware
13 Posts 6 Posters 7.8k Views
Loading More Posts
  • Oldest to Newest
  • Newest to Oldest
  • Most Votes
Reply
  • Reply as topic
Log in to reply
This topic has been deleted. Only users with topic management privileges can see it.
  • C
    Class889
    last edited by Feb 4, 2013, 7:39 PM

    I've done a test of IPERF to measure the throughput of the pfsense.
    what i've done is i entered the pfsense box via ssh and run iperf -s, then with my pc i did iperf -c pfsense.
    the way i tested does this measure wan-lan? or lan-lan? or none of them?

    well anyway i did the test for about 5 minutes and i run top -SH to see whats going on, and i saw that {idle: cpu0} was at 1.66%.
    my cpu is P4 3.06Ghz HT. (1 core, 2 threads)
    based on this picture it means the cpu is limiting my traffic?

    i got no idea what {swi:5} is.
    so i'm trying to understand if this thing limiting or the cpu.
    im using 2 TP-LINK 1Gbit network cards, 1 PCI and 1 PCIxE.

    1 Reply Last reply Reply Quote 0
    • P
      podilarius
      last edited by Feb 4, 2013, 8:11 PM

      The PCI bus will limit any connection to about 350Mbits/Sec. It is a limitation of the PCI bus and not the CPU.
      on a scale if 1 to 100 1.66 is not that much. That is unless you are talking about load average, in which case the 1.66 is interpreted as you need 1.66 number of CPUs to handle the load.
      Since you have have 2 CPUs (1 plus 1 hyperthread), you are effectively running at 83%.
      Are you running any packages or doing any VPNing?

      1 Reply Last reply Reply Quote 0
      • C
        Class889
        last edited by Feb 4, 2013, 8:34 PM Feb 4, 2013, 8:32 PM

        im not running any vpn
        the only packages i installed are: iperf, iftop, tcptraceroute, bwm-ng and nmap.

        i did how ever get to the speeds that you can see here:

        and the thing is when im not doing any test the [idle: cpu0] is at 100%
        so i think its the opposite , 100% = 100% idle, 1% = 1% idle meaning 99% stress?

        1 Reply Last reply Reply Quote 0
        • W
          wallabybob
          last edited by Feb 4, 2013, 9:55 PM

          @Class889:

          I've done a test of IPERF to measure the throughput of the pfsense.
          what i've done is i entered the pfsense box via ssh and run iperf -s, then with my pc i did iperf -c pfsense.
          the way i tested does this measure wan-lan? or lan-lan? or none of them?

          It reports throughput BETWEEN a LAN system and pfSense. This is NOT an effective way of measuring throughput when pfSense is routing.

          @Class889:

          well anyway i did the test for about 5 minutes and i run top -SH to see whats going on, and i saw that {idle: cpu0} was at 1.66%.
          my cpu is P4 3.06Ghz HT. (1 core, 2 threads)
          based on this picture it means the cpu is limiting my traffic?

          Yes.

          If you want to measure pfSense's performance as a router you need to offload the iperf server function to another system so that pfsense is forwarding traffic between the iperf server and client.

          What do you want to measure?

          1 Reply Last reply Reply Quote 0
          • R
            Roots0
            last edited by Feb 4, 2013, 10:27 PM

            @podilarius:

            The PCI bus will limit any connection to about 350Mbits/Sec. It is a limitation of the PCI bus and not the CPU.
            on a scale if 1 to 100 1.66 is not that much. That is unless you are talking about load average, in which case the 1.66 is interpreted as you need 1.66 number of CPUs to handle the load.
            Since you have have 2 CPUs (1 plus 1 hyperthread), you are effectively running at 83%.
            Are you running any packages or doing any VPNing?

            The PCI bandwidth limit is 133 MB/s is it not? So you should have no problem pushing 1Gbps in one direction.

            Does look like your cpu is limiting your throughput. Decent intel NIC's might help also using some settings tweaks. Probably best to upgrade your cpu to and i3/i5 though.

            Mobile Computer & Network Support Stockport, UK
            www.timotten.co.uk

            1 Reply Last reply Reply Quote 0
            • P
              podilarius
              last edited by Feb 4, 2013, 10:52 PM

              Based on you PIC IDLE is consuming 100% of your proc. This means nothing is going on. Please refer to your load average of 0.00. This also means nothing is going on.
              Are you new to BSD and unix based systems? I highly doubt that a hyperthreaded 3.06GHz is limiting the bandwidth that low. Most likely is the 32bit PCI bus. Which at most runs a 66MHz and the bandwidth is shared on entire bus. So if there is any other cards, you are going to slow down. I guess that is between the PCI-e and the host. Cause that won't be on the PCI NIC.
              Please reference http://en.wikipedia.org/wiki/Conventional_PCI.
              There are many other factors as well. Perhaps a screen shot of top while nearing the end of the test run would be helpful. The first screen shots looks to be under load, but you are still not using all the proc there, just 80.3%.
              Unless you can give me a screen shot with no idle and a load average over 2, I would suspect PCI bus or cable. Not to mention that of the computer you are actually using on the other side of the iperf test.
              I am not saying that CPU won't limit traffic, I just don't think that is the case here.

              This has a good write up about that also:
              http://www.enterprisenetworkingplanet.com/netsysm/article.php/3485486/Squeeze-Your-Gigabit-NIC-for-Top-Performance.htm

              It also has some interesting things you can do to speed up your network. He was testing with a P4 3Ghz which I don't think has hyperthreading.

              1 Reply Last reply Reply Quote 0
              • P
                podilarius
                last edited by Feb 4, 2013, 11:00 PM

                This is a good write up as well.
                https://calomel.org/network_performance.html

                1 Reply Last reply Reply Quote 0
                • C
                  Class889
                  last edited by Feb 5, 2013, 4:04 PM

                  thank you for all the replies :)
                  i've read one of the articles already and i know those things, but you were right that this test doesnt say anything about lan-lan throughput nor lan-wan.
                  anyway i did a test of lan-wan and i had so many issues, had massive packet loss.
                  it seems the PCIe TPLINK card is the cause, once i changed it with the OB i didnt get any packet loss and i was able to test throughput, but the other side is limiting because its a weak pc.
                  i'm gonna get some hightech laptop from a friend and ill test it.

                  my system is I7 860@ 4Ghz but the other one is E2160 @ 1.6Ghz (it gets to 90% stress while doing iperf test)
                  and my onboard card does some issues as well.
                  so ill try to find some intel server card for cheap, and get some laptop to test it, and ill update the thread.
                  once ill do that, ill see where do i stand in terms of throughput and if the P4 3.06Ghz is limiting.

                  thank you all for helping!!

                  1 Reply Last reply Reply Quote 0
                  • P
                    podilarius
                    last edited by Feb 5, 2013, 6:16 PM

                    As reference:

                    http://forum.pfsense.org/index.php/topic,53185.0.html
                    CPU in the test was a 2.13GHz processor dual core and at 939Mbits/s on a liceCD test, he was only at 30% CPU usage.

                    http://forum.pfsense.org/index.php/topic,45439.0.html
                    Same with this guy except he was pushing over 800 at 30% CPU usage.

                    Both of those are PCI-e based NICs and probably the 939MBits are Intel. Check powerD settings though before trying to go for top throughput.

                    Just saying.

                    1 Reply Last reply Reply Quote 0
                    • C
                      Class889
                      last edited by Feb 5, 2013, 7:00 PM Feb 5, 2013, 6:57 PM

                      i agree on the powerD statement but first i must know what my router is capable of :)

                      but i wonder, how come ppl dont pass the 1gbit barrier? if u're using iperf -d you should get more than 1gbit throughput max 2gbit.
                      yet i havent seen anyone pass the 1gbit barrier, and somehow i'm not able to find an answer.

                      i mean 1 pc on the lan side uploading at 1gbit and the other pc on the wan side downloading at 1gbit, combined 2gbit.
                      am i missing something? do u require that much of cpu power for such thing?

                      i mean every 1gbit card is baiscly total of 2gbit at full duplex.

                      1 Reply Last reply Reply Quote 0
                      • P
                        podilarius
                        last edited by Feb 5, 2013, 7:33 PM

                        2Gigabit is bi-directional (or full duplex if you like). Gigabit adapters can only go 1 gigabit in one direction. Most tests out there don't test them together, but one at a time. Most of the time, your traffic will be in one direct or the other, rarely will you use 1 gigabit up and down at the same time. Even in your tests, the -r test one way then the other way. I think iperf -d will do that at the same time. (source:http://doc.pfsense.org/index.php/Iperf_man_page).
                        If you took that example at 939Mbits/sec and really looked at it, one side is send while the other is receiving. While the test is one directional, you can add them together to get an effective rate. I would like to have seen his results with the -d option used. Again though, it reports in and out seperately even though calculated at the same time. So you will have to add them. Also adding threads increases throughput as well. 2 data streams is better than one.
                        There are many things that can effect top speed on a network. Generally CPU is the last to look at, that is at least with modern stuff. An 8086 could not cut it, but your 3GHz proc should without hitting the max.

                        1 Reply Last reply Reply Quote 0
                        • F
                          fragged
                          last edited by Feb 6, 2013, 7:15 AM

                          I did a test with iperf on my G630T + S1200KP <-> Desktop running i5 3570k + Asrock Z77 Pro4 (Realtek PCIe GBE family):

                          900 Mbits pf -> desktop
                          200 Mbits desktop -> pf

                          pfSense running iperf as server. I think the limiting factor here is my desktop's NIC.

                          1 Reply Last reply Reply Quote 0
                          • stephenw10S
                            stephenw10 Netgate Administrator
                            last edited by Feb 6, 2013, 8:47 AM

                            Hmm, interesting result. I wonder what is causing such a huge discrepancy?

                            @Class889:

                            i agree on the powerD statement but first i must know what my router is capable of :)

                            The P4 chip you are using does not have any useful power saving features. It would be almost pointless running powerd.

                            @Class889:

                            but i wonder, how come ppl dont pass the 1gbit barrier?

                            It's because people are, mostly, measuring the throughput of the pfSense box. Usually what people are interested in is how much bandwidth can I handle with X hardware between WAN and LAN. Testing that bi-directionally is just not relevant.

                            Steve

                            1 Reply Last reply Reply Quote 0
                            13 out of 13
                            • First post
                              13/13
                              Last post
                            Copyright 2025 Rubicon Communications LLC (Netgate). All rights reserved.
                              This community forum collects and processes your personal information.
                              consent.not_received