[c-nsp] How to measuse the throughput on Internet Links

Thu Jan 10 09:35:17 EST 2008

Having spent a fair amount of time on this type of thing recently, let me
offer a few thoughts-

With any type of speed testing, you must define the testing domain before
you start, or the numbers won't mean a thing. The reason for this is that
delay between the endpoints affects max throughput, especially with regards
to TCP.

With TCP windowing, a transmitting host can only transmit data up to the
size of the window of the receiving host before it has to stop and wait for
an ACK. This is by design as when TCP was developed, the large pipes we have
today weren't even on the drawing board.

The delay I mentioned above directly affects how long the transmitting host
sits idle waiting on an ACK, which of course affects throughput. This is why
the TCP window size should be tweaked to fit the pipe, and the max expected
delay. This is also the reason why it's rare that a single TCP session will
be able to achieve the max throughput on a given high speed pipe.

There's some good research out there on the affect of TCP Window size on
throughput. Here's a couple to get you started-
http://www-didc.lbl.gov/TCP-tuning/
http://www.psc.edu/networking/projects/tcptune/

Joe

On 1/10/08, Ramcharan, Vijay A <vijay.ramcharan at verizonbusiness.com> wrote:
>
> If you just need to do a blind bandwidth test, it's possible to use
> iperf to test your link one-way outbound without really needing a
> receiver at the other end. See below for an example. You'd just need to
> tweak the iperf settings to achieve your desired traffic rate. 10M below
> means use a 10Mbps bandwidth. I ran 2 streams which would correspond to
> the avg 20Mbps SUM bandwidth noted.
> If you have a few high speed Internet users you could probably test
> inbound as well. The test below was run from a Win2K server just behind
> the router to an IP address reachable by that router across a DS3. Note
> that iperf consumed 100% CPU on this server during the test so use
> caution when testing.
> I've also tested locally and was easily able to generate 100Mbps of
> traffic off my fairly fast PC.
>
> Since you appear to want to gauge real max bandwidth possible you'll
> probably want to try using TCP with a real receiver at the other end to
> gather stats from both the iperf sender and receiver as well as your
> router.
>
> C:\Program Files\iperf-2.0.2\bin>iperf -c 192.168.255.21 -u -b 10m -P 2
> -f k -t 600
> ------------------------------------------------------------
> Client connecting to 192.168.255.21, UDP port 5001
> Sending 1470 byte datagrams
> UDP buffer size: 63.0 KByte (default)
> ------------------------------------------------------------
> [  4] local 63.91.167.53 port 1738 connected with 192.168.255.21 port
> 5001
> [  3] local 63.91.167.53 port 1737 connected with 192.168.255.21 port
> 5001
> [  4]  0.0-600.0 sec  729444 KBytes  9959 Kbits/sec
> [  4] Sent 508130 datagrams
> [  3]  0.0-600.0 sec  731400 KBytes  9986 Kbits/sec
> [  3] Sent 509492 datagrams
> [SUM]  0.0-600.0 sec  1460844 KBytes  19945 Kbits/sec
> read failed: Connection reset by peer
> read failed: Connection reset by peer
> [  4] WARNING: did not receive ack of last datagram after 1 tries.
> [  3] WARNING: did not receive ack of last datagram after 1 tries.
>
>
>
> DS3-RTR5#sh int s2/0
> Serial2/0 is up, line protocol is up
>   Hardware is DSXPNM Serial
>   Internet address is 1.2.3.4/30
>   MTU 4470 bytes, BW 44210 Kbit, DLY 200 usec,
>      reliability 255/255, txload 107/255, rxload 1/255
>   Encapsulation HDLC, crc 16, loopback not set
>   Keepalive set (10 sec)
>   ...
>   5 minute input rate 48000 bits/sec, 68 packets/sec
>   5 minute output rate 18560000 bits/sec, 3067 packets/sec
> ...
> DSU mode 0, bandwidth 44210, real bandwidth 44210, scramble 0
> DS3-RTR5#
>
> Vijay Ramcharan
>
> -----Original Message-----
> From: cisco-nsp-bounces at puck.nether.net
> [mailto:cisco-nsp-bounces at puck.nether.net] On Behalf Of omar parihuana
> Sent: January 09, 2008 11:57
> To: Michael Long; cisco-nsp at puck.nether.net
> Subject: Re: [c-nsp] How to measuse the throughput on Internet Links
>
> Hi Michael,
>
> thanks for your response....
>
> I tried to use IPERF previously without success because set up the
> server
> was difficult for me. I'm looking for other options.
>
> Thanks again
> Rgds.
>
>
> On 1/9/08, Michael Long <mlong at mikesoffice.org> wrote:
> >
> > iperf. http://dast.nlanr.net/Projects/Iperf/
> >
> > Although you'll need a client and server setup. Plus it helps if the
> > server is somewhat close to the link you are trying to test.
> >
> > Mike
> >
> > omar parihuana wrote:
> > > Hi guys,
> > >
> > > I bought a new Internet Link 40Mbps 1:1 that is provide with
> > FastEthernet
> > > interface, now, I would like to check if I have the 40Mbps
> effective.
> > How
> > > can I measure the max capacity of my Internet link? any suggestions?
> > >
> > > Thanks in advanced...
> > >
> > > Rgds.
> > >
> >
>
>
>
> --
> Omar E.P.T
> -----------------
> Certified Networking Professionals make better Connections!
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>