Anyone know if ttcp should work in Cmode?

[ryan_foster]

Hi,

Were implementing a new cluster and investigating a potential performance problem with the network and need to test the network throughput between a RedHat server and a interface within the cluster.

In 7-Mode ttcp could be used for this and the command still exists within the nodeshell but i can't get it to work in cmode as it can't establish a connection between the ttcp sender and receiver. 

Command on Netapp Node

xxcluster01> run -node xxcluster01-01

xxcluster01-01> priv set diag

xxcluster01-01*> ttcp -rs

ttcp -r : nbuf=1024, buflen=1024, sock_buf=262144 port=5001

ttcp-r: socket

Command on Redhat Client:

ttcp -t -s -f g -n 409600 -p 5001 10.18.40.251
ttcp-t: buflen=8192, nbuf=409600, align=16384/0, port=5001 tcp -> 10.18.40.251
ttcp-t: socket
ttcp-t: connect: Connection refused
errno=111

I'm thinking that although the command still exists it may not be possible to use as the interfaces are configured within the clustershell and the nodeshell therefore has no route to the network.   Although hopefully i'm doing something wrong or missed a step to get it to work in cmode.  Any idea?

[ryan_foster]

Answered my own question if anyone needs this in the future.    Using priv set diag on the nodeshell allows you to configure an IP address on the interface you want to test, you then just need to specify that IP on the ttcp sender.

[amt]

Hi Ryan ,

Can you let me know how it worked.

I have my setup  FILER (Cmode) <----> SWITCH <------> LANFORGE ( Federo client ).

Configured ip address on filer port and lanforge in same subnet. Ping is successfull.

First ran - > ttcp -r -s on Filer and then from lanforge ran - >  ttcp -t -s -n 40000000 -l 10000 -p 5001 < Filer ip address > .

But getting connection refused error.

[LANFORGE-1 ~]# ttcp -t -s -n 40000000 -l 100000 -p 5001  200.20.20.1

ttcp-t: buflen=100000, nbuf=40000000, align=16384/0, port=5001  tcp  -> 200.20.20.1

ttcp-t: socket

ttcp-t: connect: Connection refused

errno=111

If i have done anything wrong or let me know what i need to do.

Thanks ,

Amit

[ryan_foster]

Hi Amit,

The command looks correct my only guess is that you have setup the interface on the cluster/vserver rather than one of the physical nodes itself.   ttcp is only available on the physical nodes so needs an interface setup on same IP range as server using priv set diag in the nodeshell.   Once setup just run the following.

Start receiver on Filer:

- run -node nodemane

- priv set diag

- ttcp -rs

Start Sender on Server to IP you setup on the physical node

- ttcp -s -t -n 409600  ipaddress

Make sure the ttcp ports are the same, by default i think it uses 5001

[LarsSchneider]

Hi,

thanks Ryan for sharing your solution. Even though I would like to reopen this topic because the solution which has been described is a "quick and dirty"-fix. Adding another interface to the physical node seems to me tolerable if you are working in a test environment or something similar. But for testing the tcp throughput from an actual productive system to another host/client it's definitly inadequate. Therefore I would like to ask: "How do you test in cMod the TCP throughput from one filer to an host?" I'm using iperf on the host to generate the data. In 7-Mode I used the command "ttcp -s" (click) to use the filer as a sink. Again, it's important for me to get a "clean" solution without creating an interface on the physical node.

Any help is appreciated and thanks in advance.

Greetings

Lars Schneider