-5

Is it possible to test a NIC latency by looping ports together? Can it be done at least in theory? Will ethernet work between 2 ports of the same card? or between NIC port and LOM port?

preferably on Windows 2008 R2

1 Answer 1

2

You can loop them, but be sure you don't have setup like NIC teaming or something else.

All modern NIC can be connected using straight cable, but if it's oldest one you need to use cross cable.

Before this, check manufacturer site, most of them has tools (like Intel, Broadcom) which can make loopback test (and lot of others) without any special conditions like this.

In real, I don't think that you can measure useful values when you make loop like nullmodem on 2 local ethernet ports.

Latency is basically caused by used bandwidth/traffic on L2 switches, so making local loopback test makes no sense for latency measurement.

7
  • I was thinking of running 2 apps. one will be sending and the other will be listening. hence i could figure out latency of the card.... what do u think? Jun 12, 2012 at 4:39
  • No. 99% of latency issues caused by traffic.. What I need is to measure latency of the NIC itself (and OS layers on top of it)... I am that 1% ;) who doesnt care about switches and cables etc. Jun 12, 2012 at 5:14
  • @Bobb all your traffic can by shaped/dealyed on switch using QoS, so testing it locally does not reflect real environment usage.I would recommend also use performance counters & monitoring tools of win2k8 to be involved into your tests.
    – user106666
    Jun 12, 2012 at 5:17
  • 1
    It is very simple really. latency L = A + B + ... + X. You never probably looked at X (which is NIC, driver and OS stack). Because X usually below 1ms. However X is still there. Solving A B etc will not remove X.. What you are saying is that - you dont see how it could ever be important.. thats all.. now imagine that I am not just a bored idiot and I have an agenda. and in that agenda - X is very important. It is that important that I want to pay $1000 for a NIC and $2000 for a server which can bring X down from 50 microseconds to 10.... can you accept that? Jun 12, 2012 at 5:54
  • @Bobb 1+, now fully understand your concept, maybe you should update your question, because this is key point as I understand now
    – user106666
    Jun 12, 2012 at 5:57

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .