NDT-DEV email list created

Status: New
Owner: ----
Labels: Type-Defect Severity-Medium

New issue 73 by : Running ndt web100srv without dns/  
etc/hosts entries take an unnecessarily long time - tests sometimes fail
http://code.google.com/p/ndt/issues/detail?id=73

What steps will reproduce the problem?
1.Run ndt without a dns server or any entries in /etc/hosts for client or  
0.0.0.0
2.Connect with a client and you will see pauses between tests
3.The accumulated delay can be be enough to make all the test timeout and  
fail completely.

What is the expected output? What do you see instead?
Tests run fully and completely without delays/ timeouts.
Client failure upon timeout looks like this
~~~
[root@machine3
 ~]# time web100clt -n 172.20.1.2 -ll
Testing network path for configuration and performance problems  --  Using  
IPv4 address
Checking for Middleboxes . . . . . . . . . . . . . . . . . .  Done
checking for firewalls . . . . . . . . . . . . . . . . . . .  Done
running 10s outbound test (client to server) . . . . .  941.55 Mb/s
running 10s inbound test (server to client) . . . . . . Protocol error -  
missed text message!
S2C throughput test FAILED!
Protocol error - missed prepare message!
META test FAILED!
Protocol error - expected results!  got '', msgtype=2

real    2m8.677s
user    0m0.392s
sys     0m3.194s
~~~

As opposed to what it should look like

~~~
[root@machine3
 ~]# time web100clt -n 172.20.1.2 -ll
Testing network path for configuration and performance problems  --  Using  
IPv4 address
Checking for Middleboxes . . . . . . . . . . . . . . . . . .  Done
checking for firewalls . . . . . . . . . . . . . . . . . . .  Done
running 10s outbound test (client to server) . . . . .  941.55 Mb/s
running 10s inbound test (server to client) . . . . . . 922.28 Mb/s
sending meta information to server . . . . . Done

... results here ...

real    0m32.601s
user    0m0.858s
sys     0m5.643s
~~~

Even running with the clients IP in /etc/hosts but not 0.0.0.0 takes about  
50 seconds for a test to complete successfully.

What version of the product are you using? On what operating system?
Debian wheezy Linux, ndt build based on latest svn r796.

Please provide any additional information below.
This is caused by the I2util library looking up every(?) connection with  
getnameinfo, running server on debian wheezy I experience a 5 second  
timeout before missing entries fall back to IP address.

The quick fix for this is to add the NI_NUMERICSERV flag to the I2Util  
getnameinfo call. I don't feel that resolving hostnames is that important  
especially as most of the time they are not being printed to the user.

Attached are server output running debug level 6 with times to complete the  
_I2AddrSetNodePort() (which does the getnameinfo call) function logged.  
i2MONO diff: is the real time - i2CPU diff: is CPU time - NAME IS: is the  
name it resolved to.

Attachments:
        Without DNS lookup attempting to resolve  12.8 KB
        With NI_NUMERICSERV flag  20.6 KB