2

I have a Debian box up and running with KVM and two guest machines, which I will refer here as Tux-A and Tux-B, both bind to the same physical network card of the Host via a bridged connection.

The problem is that Tux-A shows evidences of some sort of bandwidth limit uploading to the outer network:

  • upload from Tux-A to ip on same Private Network = 45Mb/sec
  • upload from Tux-A to ip on outer Public Network = 150Kb/sec
  • upload from Tux-B to ip on same Private Network = 45Mb/sec
  • upload from Tux-B to ip on outer Public Network = 3Mb/sec

given these values I presume that:

  • no limit is set on top via KVM to the upload bandwidth (otherwise Tux-B should also be slow uploading to public network)
  • no limit is set on top of Tux-A (otherwise Tux-A should also be slow uploading to private network)

just to complete the scenario:

  • no iptables is active on the Debian host nor Tux-A (double checked and on both machines the services are disabled)
  • no other specific service in terms of traffic shaping is active on both the Debian host and Tux-A (no tc, no shorewall, no KVM parameters tweaking, nothing)
  • the router that connects to outer network is (said to be) uncapable of any form of traffic shaping (and I have no evidence of that capability in the admin area)

So in my (very very humble) opinion, if it's something set directly on Tux-A, and if on Linux everything is a file, is there somewhere among Linux's system files where this kind of limit can be set? somewhere in /proc, /sys, or such? or can someone suggest some other way to solve this situation?

Improve this question

1 Answer 1

Reset to default
1

It is not probably traffic shapping.

A emulated VM is not able to generated too much traffic if the NIC is emulated instead of paravirtualized. (like 10 times less traffic when emulated)

Install the virtIO driver and select a virtio card on the slowest machine.

Improve this answer
2
  • I'll try your suggestion, but how do you explain that uploading from Tux-A to local network goes full speed? and also why Tux-B does not suffer the same problem (while being configured the same way as Tux-A with no virtio card)?
    – funkoolow
    Jul 29, 2018 at 14:21
  • Lots of unknown variables here. Obviously, knowing your network I cannot comment, however I do have seen up to a 10x improvement in the past in some situations. Nevertheless, as you are confirming you are not using paravirtualized interfaces, I do advise it, as besides possible performance improvements, you are saving CPU load. Mind you that it would be more interesting commenting after trying it and not before.
    – Rui F Ribeiro
    Jul 29, 2018 at 17:14

You must log in to answer this question.

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