Mailing List for the Bandwidth Testing Tool Iperf

Aaron Brown | 16 Feb 2010 14:55

Re: Iperf UDP Packet Loss

On Feb 14, 2010, at 8:51 PM, Wichai Komentrakarn wrote:

I am trying to use the Iperf to analyze UDP packet loss on a network. I saw Iperf reported several % packet loss but when I used Wireshark to capture the sent and received packets and compared them. I couldn't see any packets loss on the received side. Can you explain how Iperf reported percentage of packet loss or what is the condision that Iperf decided it didn't recieve a UDP packet.

Since UDP is a lossy protocol, the kernel does not guarantee that it will send data its been handed, and will sometimes drop the packets if, for example, the send or receive buffer are filled. It could also be that the packets got corrupted on the wire, didn't checksum properly and were tossed out.

Cheers,

Aaron

Internet2 Spring Member Meeting
April 26-28, 2010 - Arlington, Virginia
http://events.internet2.edu/2010/spring-mm/

------------------------------------------------------------------------------
SOLARIS 10 is the OS for Data Centers - provides features such as DTrace,
Predictive Self Healing and Award Winning ZFS. Get Solaris 10 NOW
http://p.sf.net/sfu/solaris-dev2dev

_______________________________________________
Iperf-users mailing list
Iperf-users@...
https://lists.sourceforge.net/lists/listinfo/iperf-users

headers

Feghali, Jose | 16 Feb 2010 18:18

Re: Iperf UDP Packet Loss

Aaron,

That may well be the case – but FWIW, when I was testing DVTS with the University of São Paulo last year, we consistently got huge packet loss reported by iperf depending on the settings and which Iperf version we used. Linux iperf versions did not give the same results. I compiled a newer version of Iperf for use with Windows which “behaved” better, but only after adjusting the –w settings. It consistently reported out-of-order packets which Wireshark and other sniffers didn’t “see” – and no artifact was seen on DVTS. I believe we communicated about this then, but no clear explanation could be found for that behavior – at least not that I recollect. There are, of course, a number of parameters that could affect iperf but we did try to have both computers network details matched – and tried several network cards.

All best,

José

From: Aaron Brown [mailto:aaron-H4aWS73dXup+qImEYqgU8Q@public.gmane.org]
Sent: Tuesday, February 16, 2010 7:56 AM
To: Wichai Komentrakarn
Cc: iperf-users-IlAC/IgH3vHkIehEPBwdNw@public.gmane.org
Subject: Re: [Iperf-users] Iperf UDP Packet Loss

On Feb 14, 2010, at 8:51 PM, Wichai Komentrakarn wrote:

I am trying to use the Iperf to analyze UDP packet loss on a network. I saw Iperf reported several % packet loss but when I used Wireshark to capture the sent and received packets and compared them. I couldn't see any packets loss on the received side. Can you explain how Iperf reported percentage of packet loss or what is the condision that Iperf decided it didn't recieve a UDP packet.

Cheers,

Aaron

Internet2 Spring Member Meeting
April 26-28, 2010 - Arlington, Virginia
http://events.internet2.edu/2010/spring-mm/

------------------------------------------------------------------------------
SOLARIS 10 is the OS for Data Centers - provides features such as DTrace,
Predictive Self Healing and Award Winning ZFS. Get Solaris 10 NOW
http://p.sf.net/sfu/solaris-dev2dev

_______________________________________________
Iperf-users mailing list
Iperf-users@...
https://lists.sourceforge.net/lists/listinfo/iperf-users

headers

Marc Herbert | 16 Feb 2010 20:07

Re: Iperf UDP Packet Loss

2010/2/16 Feghali, Jose <j.feghali@...>:

> That may well be the case – but FWIW, when I was testing DVTS with the
> University of São Paulo last year, we consistently got huge packet loss
> reported by iperf depending on the settings and which Iperf version we used.
> Linux iperf versions did not give the same results. I compiled a newer
> version of Iperf for use with Windows which “behaved” better, but only after
> adjusting the –w settings.

The socket buffer size has an impact on how some kernels drop UDP
packets. You will find more information about this in the archive of
this list.

------------------------------------------------------------------------------
SOLARIS 10 is the OS for Data Centers - provides features such as DTrace,
Predictive Self Healing and Award Winning ZFS. Get Solaris 10 NOW
http://p.sf.net/sfu/solaris-dev2dev
_______________________________________________
Iperf-users mailing list
Iperf-users@...
https://lists.sourceforge.net/lists/listinfo/iperf-users

headers

Feghali, Jose | 17 Feb 2010 19:33

Re: Iperf UDP Packet Loss

Marc,

Thanks for the note. 

For clarification - the main reason we decided to stop investigating this problem was that later on we had
the New World Symphony as a second testing site for comparison purposes. With the same version of iperf on
all computers, the results with the NWS were perfect, while the ones with USP were bad (although the
streams we sent and received perfectly). Obviously something about the link with Brazil was interfering
with the iperf results, but it worked just fine with Miami. Maybe iperf didn't like the router setup in
Brazil...   

-----Original Message-----
From: marc.herbert@...
[mailto:marc.herbert@...] On Behalf Of Marc Herbert
Sent: Tuesday, February 16, 2010 1:07 PM
To: Feghali, Jose
Cc: iperf-users@...
Subject: Re: [Iperf-users] Iperf UDP Packet Loss

2010/2/16 Feghali, Jose <j.feghali@...>:

> That may well be the case - but FWIW, when I was testing DVTS with the
> University of São Paulo last year, we consistently got huge packet loss
> reported by iperf depending on the settings and which Iperf version we used.
> Linux iperf versions did not give the same results. I compiled a newer
> version of Iperf for use with Windows which "behaved" better, but only after
> adjusting the -w settings.

The socket buffer size has an impact on how some kernels drop UDP
packets. You will find more information about this in the archive of
this list.

------------------------------------------------------------------------------
SOLARIS 10 is the OS for Data Centers - provides features such as DTrace,
Predictive Self Healing and Award Winning ZFS. Get Solaris 10 NOW
http://p.sf.net/sfu/solaris-dev2dev
_______________________________________________
Iperf-users mailing list
Iperf-users@...
https://lists.sourceforge.net/lists/listinfo/iperf-users

headers

Yousuf Ahmad | 22 Feb 2010 22:24

Iperf UDP parameters

Hello,

I am running a test on two windows machines that are connected to each other through a local Gigabit 8-port switch. One is a Dell power edge Windows 2003 server and one is an XP laptop. Ports on both machines are running at 1Gbps and they have no other apps running on them besides iperf. Windows 2003 server is acting as an iperf server and the laptop is acting as client

When I run udp test using ‘-b100m’ on the client with server running in default, I get about 95Mbps throughput but with lots of dropped packets. I then start to increase the buffer size on both ends and find that at buffer size of ‘-w32MB’ dropped packets get down to zero. To find max.udp throughput on the 1Gbps link, I increased the bandwidth gradually to ‘the max. of –b2147m’ and found that the client would not even send anything faster than 115Mbps regardless of what I set the buffer size to be. It was only when I decreased the datagram size to ‘-l900B’ on both ends that the client started sending data at about 700Mbps (even then it was inconsistent –kept varying from 500-800Mbps-).

How does iperf use “b, w and l” parameters? Is there a way I can calculate or estimate their value in order to achieve max. UDP throughput for a given link?

Below are some samples from my test (server reports for each are at the bottom):

Thank you,

Yousuf,

C:\Iperf>iperf -c10.10.10.80 -u -i1 -b100m -w16KB

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

Client connecting to 10.10.10.80, UDP port 5001

Sending 1470 byte datagrams

UDP buffer size: 16.0 KByte

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

[1912] local 10.10.10.81 port 1680 connected with 10.10.10.80 port 5001

[ ID] Interval Transfer Bandwidth

[1912] 0.0- 1.0 sec 10.9 MBytes 91.2 Mbits/sec

[1912] 1.0- 2.0 sec 11.8 MBytes 98.9 Mbits/sec

[1912] 2.0- 3.0 sec 11.8 MBytes 98.9 Mbits/sec

[1912] 3.0- 4.0 sec 11.8 MBytes 98.9 Mbits/sec

[1912] 4.0- 5.0 sec 11.6 MBytes 97.4 Mbits/sec

[1912] 5.0- 6.0 sec 11.4 MBytes 95.8 Mbits/sec

[1912] 6.0- 7.0 sec 11.1 MBytes 92.7 Mbits/sec

[1912] 7.0- 8.0 sec 11.2 MBytes 94.3 Mbits/sec

[1912] 8.0- 9.0 sec 11.4 MBytes 95.8 Mbits/sec

[1912] 9.0-10.0 sec 11.2 MBytes 94.3 Mbits/sec

[1912] 0.0-10.0 sec 114 MBytes 95.7 Mbits/sec

[1912] Server Report:

[1912] 0.0-10.0 sec 112 MBytes 93.8 Mbits/sec 0.000 ms 1882/81491 (2.3%)

[1912] Sent 81491 datagrams

C:\Iperf>iperf -c10.10.10.80 -u -i1 -b100m -w1024KB

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

Client connecting to 10.10.10.80, UDP port 5001

Sending 1470 byte datagrams

UDP buffer size: 1.00 MByte

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

[1912] local 10.10.10.81 port 1737 connected with 10.10.10.80 port 5001

[ ID] Interval Transfer Bandwidth

[1912] 0.0- 1.0 sec 11.2 MBytes 94.3 Mbits/sec

[1912] 1.0- 2.0 sec 11.4 MBytes 95.8 Mbits/sec

[1912] 2.0- 3.0 sec 11.5 MBytes 96.9 Mbits/sec

[1912] 3.0- 4.0 sec 11.7 MBytes 97.9 Mbits/sec

[1912] 4.0- 5.0 sec 12.0 MBytes 100 Mbits/sec

[1912] 5.0- 6.0 sec 11.8 MBytes 98.9 Mbits/sec

[1912] 6.0- 7.0 sec 11.8 MBytes 98.9 Mbits/sec

[1912] 7.0- 8.0 sec 11.4 MBytes 95.8 Mbits/sec

[1912] 8.0- 9.0 sec 11.2 MBytes 94.3 Mbits/sec

[1912] 9.0-10.0 sec 11.2 MBytes 94.3 Mbits/sec

[1912] 0.0-10.0 sec 115 MBytes 96.6 Mbits/sec

[1912] Server Report:

[1912] 0.0-10.0 sec 115 MBytes 96.8 Mbits/sec 0.000 ms 0/82287 (0%)

[1912] Sent 82287 datagrams

C:\Iperf>iperf -c10.10.10.80 -u -i1 -b2147m -w32MB

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

Client connecting to 10.10.10.80, UDP port 5001

Sending 1470 byte datagrams

UDP buffer size: 32.0 MByte

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

[1912] local 10.10.10.70 port 2323 connected with 10.10.10.80 port 5001

[ ID] Interval Transfer Bandwidth

[1912] 0.0- 1.0 sec 13.8 MBytes 116 Mbits/sec

[1912] 1.0- 2.0 sec 14.0 MBytes 117 Mbits/sec

[1912] 2.0- 3.0 sec 14.2 MBytes 119 Mbits/sec

[1912] 3.0- 4.0 sec 14.3 MBytes 120 Mbits/sec

[1912] 4.0- 5.0 sec 14.0 MBytes 118 Mbits/sec

[1912] 5.0- 6.0 sec 14.2 MBytes 119 Mbits/sec

[1912] 6.0- 7.0 sec 13.6 MBytes 114 Mbits/sec

[1912] 7.0- 8.0 sec 13.6 MBytes 114 Mbits/sec

[1912] 8.0- 9.0 sec 13.8 MBytes 115 Mbits/sec

[1912] 9.0-10.0 sec 13.8 MBytes 116 Mbits/sec

[1912] 0.0-10.0 sec 139 MBytes 117 Mbits/sec

[1912] Server Report:

[1912] 0.0-10.0 sec 139 MBytes 117 Mbits/sec 1.201 ms 0/99331 (0%)

[1912] Sent 99331 datagrams

C:\Iperf>iperf -c10.10.10.80 -u -i1 -b2147m -w32MB -l900B

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

Client connecting to 10.10.10.80, UDP port 5001

Sending 900 byte datagrams

UDP buffer size: 32.0 MByte

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

[1912] local 10.10.10.70 port 2325 connected with 10.10.10.80 port 5001

[ ID] Interval Transfer Bandwidth

[1912] 0.0- 1.0 sec 80.5 MBytes 675 Mbits/sec

[1912] 1.0- 2.0 sec 101 MBytes 844 Mbits/sec

[1912] 2.0- 3.0 sec 99.6 MBytes 836 Mbits/sec

[1912] 3.0- 4.0 sec 104 MBytes 869 Mbits/sec

[1912] 4.0- 5.0 sec 102 MBytes 854 Mbits/sec

[1912] 5.0- 6.0 sec 104 MBytes 873 Mbits/sec

[1912] 6.0- 7.0 sec 93.1 MBytes 781 Mbits/sec

[1912] 7.0- 8.0 sec 102 MBytes 852 Mbits/sec

[1912] 8.0- 9.0 sec 93.6 MBytes 785 Mbits/sec

[1912] 9.0-10.0 sec 77.9 MBytes 653 Mbits/sec

[1912] 0.0-10.2 sec 956 MBytes 785 Mbits/sec

[1912] Server Report:

[1912] 0.0-10.2 sec 956 MBytes 787 Mbits/sec 19.225 ms 0/1114043 (0%)

[1912] Sent 1114043 datagrams

Your E-mail and More On-the-Go. Get Windows Live Hotmail Free. Sign up now.

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev

_______________________________________________
Iperf-users mailing list
Iperf-users@...
https://lists.sourceforge.net/lists/listinfo/iperf-users

headers

Marc Herbert | 23 Feb 2010 14:00

Re: Iperf UDP parameters

2010/2/22 Yousuf Ahmad <yousufa@...>:
> How does iperf use “b, w and l” parameters?

AFAIK Iperf does nothing with the socket buffer size argument ('-w')
but pass it to your operating system.

> Is there a way I can calculate or estimate their value in order to achieve max. UDP throughput for a given link?

Not without understanding the TCP/IP stack of your operating system
(this is of course much easier with an open-source system).

Regards,

Marc

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_______________________________________________
Iperf-users mailing list
Iperf-users@...
https://lists.sourceforge.net/lists/listinfo/iperf-users

headers

Yousuf Ahmad | 23 Feb 2010 16:28

Re: Iperf UDP parameters

Thank you very much Marc,

Is there a reason why the client would not send udp data higher than ~115Mbps till I lower the datagram size to about 900 bytes? (when I do this it goes to about 700Mbps on a 1Gbps link). This is regardless of my buffer size on both the client and sever machine.

On a seperate test, I also noticed that if I have the exact same setup with identical Windows PC's connected locally to a 100Mbps switch and all my ports were running at 100Mbps, the bandwidth the client sends does not seem to follow any specific trend. For example, if I set the client to send in 1,25, or 50Mbps, the client does seem to send packets at that rate. However, when I increase it to 100Mbps, it no longer scales and continues to send data at ~50Mbps. The rate it sends the data starts to increase at about '-b150m'. It peaks at about -b185m where it sends data at about ~94Mbps. If I increase the value of '-b' a little further, the client sharply drops its sending rate to about ~63-64Mbps (again this is regardless of the buffer size I have on the client and server).

Any input you can provide on the machanics of this is greatly appreciated,

Best Regards,
Yousuf,

> Date: Tue, 23 Feb 2010 13:00:53 +0000
> Subject: Re: [Iperf-users] Iperf UDP parameters
> From: Marc.Herbert-GANU6spQydw@public.gmane.org
> To: yousufa-PkbjNfxxIARBDgjK7y7TUQ@public.gmane.org
> CC: iperf-users <at> dast.nlanr.net
>
> 2010/2/22 Yousuf Ahmad <yousufa <at> hotmail.com>:
> > How does iperf use “b, w and l” parameters?
>
> AFAIK Iperf does nothing with the socket buffer size argument ('-w')
> but pass it to your operating system.
>
>
> > Is there a way I can calculate or estimate their value in order to achieve max. UDP throughput for a given link?
>
> Not without understanding the TCP/IP stack of your operating system
> (this is of course much easier with an open-source system).
>
> Regards,
>
> Marc

Hotmail: Free, trusted and rich email service. Get it now.

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev

_______________________________________________
Iperf-users mailing list
Iperf-users@...
https://lists.sourceforge.net/lists/listinfo/iperf-users