perfSONAR User Q&A and Other Discussion

[Chad Kotil <>]

That sounds good to me.
I've submitted Enhancement #931.

https://code.google.com/p/perfsonar-ps/issues/detail?id=931

Thanks,
--Chad

On 6/13/14, 12:44 PM, Aaron Brown wrote:
> Hey Chad,
> 
> Looking through the generator, it doesn’t take the test parameters into 
> account, just that there are tests of that type between the two hosts (on 
> the theory that folks would have one test of each type between two hosts) . 
> The best bet is probably to put an issue into the tracker to request a 
> “strict” generation mode that puts in the test parameters to be able to 
> differentiate those kinds of different tests.
> 
> Cheers,
> Aaron
> 
> On Jun 12, 2014, at 1:29 PM, Chad Kotil 
> <>
>  wrote:
> 
>> Hi Andy and Aaron,
>> I'm still unable to get maddash to display the correct data from both of
>> my meshes that use a unique testspec with different parameters.
>>
>> I've generated a mesh config json file from the example here:
>> /opt/perfsonar_ps/mesh_config/doc/example.conf Then ran build_json. This
>> worked really well. From this I generated a new maddash.yaml using
>> generate_gui_configuration and even a owmesh.conf with
>> generate_configuration on an agent node. However I'm hitting the same
>> issue. Without the knowledge of the testpsec_id the data that maddash is
>> displaying is not correct. owmesh.conf knows about the different
>> testspecs, as defined here (in my Config::General format config):
>>
>> <test_spec bwctl_1h_tcp_core>
>>  type              perfsonarbuoy/bwctl  # Perform a bwctl test (i.e.
>> achievable bandwidth)
>>  tool              bwctl/iperf          # Use 'iperf' to do the
>> bandwidh test
>>  protocol          tcp                  # Run a TCP bandwidth test
>>  interval          3600                # Run the test every hour
>>  duration          30                   # Perform a 30 second test
>>  random_start_percentage     25
>>  report_interval   2
>> </test_spec>
>>
>> <test_spec bwctl_1h_udp_core>
>>  # Define a test spec for testing UDP bandwidth once every hour
>>
>>  type              perfsonarbuoy/bwctl
>>  tool              bwctl/iperf
>>  protocol          udp                  # Run a UDP bandwidth test
>>  interval          3600                # Run the test every hour
>>  duration          30                   # Perform a 30 second test
>>  udp_bandwidth     800000000             # Perform a 800Mbps test
>>  buffer_length     35896
>>  window_size       64m
>>  random_start_percentage      25
>>  report_interval   1
>> </test_spec>
>>
>> and another testepc for udp, notice the different parameters.
>> owmesh.conf has these same params and testpecids, so collecting the data
>> is OK..
>>
>> <test_spec bwctl_1h_udp_edge>
>>  # Define a test spec for testing UDP bandwidth once every hour
>>
>>  type              perfsonarbuoy/bwctl
>>  tool              bwctl/iperf
>>  protocol          udp                  # Run a UDP bandwidth test
>>  interval          3600                # Run the test every hour
>>  duration          30                   # Perform a 30 second test
>>  udp_bandwidth     200000000             # Perform a 800Mbps test
>>  buffer_length     35896
>>  window_size       64m
>>  random_start_percentage      25
>>  report_interval   1
>> </test_spec>
>>
>>
>> However , in maddash.yaml for each of the meshes i've defined, the
>> service check commands and the graphURL commands are the same. This
>> causes unexpected results in the data that is returned. You never know
>> which testspec your're going to get. (I also didnt set my thresholds
>> right, but that's another issue ill clean up later)
>>
>>    id: Bandwidth_Mesh-CORE_UDP_-_Throughput
>>    name: Throughput
>>    ok_description: Throughput >= 900Mbps
>>    params:
>>      command: '/opt/perfsonar_ps/nagios/bin/check_throughput.pl -u
>> %maUrl -w 0.9: -c 0.5: -r 86400 -s %row -d %col'
>>      graphUrl:
>> http://localhost/serviceTest/bandwidthGraph.cgi?url=%maUrl&dst=%col&src=%row&length=2592000
>>
>>    id: Bandwidth_Mesh_-EDGE_UDP_-_Throughput
>>    name: Throughput
>>    ok_description: Throughput >= 900Mbps
>>    params:
>>      command: '/opt/perfsonar_ps/nagios/bin/check_throughput.pl -u
>> %maUrl -w 0.9: -c 0.5: -r 86400 -s %row -d %col'
>>      graphUrl:
>> http://localhost/serviceTest/bandwidthGraph.cgi?url=%maUrl&dst=%col&src=%row&length=2592000
>>
>> If the issue still is not clear, look at my groups, I think this really
>> helps to explain the issue. Notice the overlap, and how node d and e
>> exist in both meshes. The UDP parameters are different. Core has a udp
>> bandwidth of 800M and edge has 200M.
>>
>> <group bwctl_udp_core>
>>  type              mesh
>>  member            bwctl.a.net
>>  member            bwctl.b.net
>>  member            bwctl.c.net
>>  member            bwctl.d.net
>>  member            bwctl.e.net
>> </group>
>> <group bwctl__udp_edge>
>>  type              mesh
>>  member            bwctl.d.net
>>  member            bwctl.e.net
>>  member            bwctl.f.net
>>  member            bwctl.g.net
>> </group>
>>
>> Then the test definitions:
>>
>> <test>
>>  description       Bandwidth CORE UDP
>>  group             bwctl_udp_core
>>  test_spec         bwctl_1h_udp_core
>> </test>
>>
>> <test>
>>  description       Bandwidth EDGE UDP
>>  group             bwctl__udp_edge
>>  test_spec         bwctl_1h_udp_edge
>> </test>
>>
>> Any ideas how I might get this to work? Do we need to make some feature
>> requests for maddash and the config gen scripts?
>>
>> Let me know if you need more info.
>>
>> Thanks,
>> --Chad
>>
>>
>> On 6/11/14, 12:30 PM, Andrew Lake wrote:
>>> Hi Chad,
>>>
>>> Just to supplement what Aaron sent, this document also has some details 
>>> on specifically using MaDDash with the mesh configuration software: 
>>> http://code.google.com/p/perfsonar-ps/wiki/MaDDashInstall#Advanced_Topic:_Using_the_perfSONAR_Mesh_Configuration_Software
>>>
>>> One thing we will likely need to add is some smarts to the mesh config to 
>>> add on the "-p" option to check_throughput.pl so it only grabs udp tests, 
>>> but that's an easy change. We may also need to add an option to filter 
>>> based on the UDP bandwidth rate, but that is also a relatively trivial 
>>> addition. I think as long as we have a fine-grained enough set of 
>>> filters, you can get away without explicitly setting a test ID. 
>>>
>>> Thanks,
>>> Andy
>>>
>>>
>>>
>>> On Jun 11, 2014, at 11:23 AM, Chad Kotil 
>>> <>
>>>  wrote:
>>>
>>>> OK Great. I've actually used this before, but never in creating my own.
>>>> I will give it a shot and let you know how it goes.
>>>> It does look like I can define multiple meshes with different parameters
>>>> for defining UDP tests.
>>>>
>>>> Thanks,
>>>> --Chad
>>>>
>>>>
>>>> On 6/11/14, 11:20 AM, Aaron Brown wrote:
>>>>> Hey Chad,
>>>>>
>>>>> The plan is to replace the owmesh configuration with the mesh 
>>>>> configuration stuff. You can get a (broad) overview of it at 
>>>>> https://code.google.com/p/perfsonar-ps/wiki/MeshConfiguration along 
>>>>> with a pointer an example configuration. If there’s some property in 
>>>>> your current meshes that isn’t easily translatable into the new mesh 
>>>>> configuration, we may be able to make some changes to better support it.
>>>>>
>>>>> Cheers,
>>>>> Aaron
>>>>>
>>>>> On Jun 11, 2014, at 10:47 AM, Chad Kotil 
>>>>> <>
>>>>>  wrote:
>>>>>
>>>>>> Ah good to know about the fate of owmesh.conf.
>>>>>
>>>>>>
>>>>>> The reason we have multiple meshes are because of the UDP tests mainly.
>>>>>> So that we can define slower speed UDP tests for some hosts. Some of 
>>>>>> our
>>>>>> hosts have a fast backbone connection and others are very slow. To 
>>>>>> avoid
>>>>>> causing congestion on the backbones we chose to create these separate
>>>>>> meshes with new testspecs. How can I accomplish this without the use of
>>>>>> owmesh.conf and unique testspecs?
>>>>>>
>>>>>> Another complicating issue im seeing in maddash with my current setup 
>>>>>> is
>>>>>> that there is overlap where a host exist in multiple meshes. Since
>>>>>> maddash and the checks are unaware of the tspec_id data can now come
>>>>>> from any of the meshes.
>>>>>>
>>>>>> --Chad
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 6/11/14, 9:49 AM, Andrew Lake wrote:
>>>>>>> Hi Chad,
>>>>>>>
>>>>>>> There are no plans to add testspec_id to maddash because the 
>>>>>>> owmesh.conf file (and thus the spec id) is going away completely in 
>>>>>>> 3.4. You can already define multiple maddash "checks" that filter on 
>>>>>>> different test parameters though. For example, for throughput you 
>>>>>>> could have two grids with the exact same hosts but one is for UDP 
>>>>>>> tests and another is for TCP tests. In the maddash check definition 
>>>>>>> you'd just need to define a "command" option in one that has "-p tcp" 
>>>>>>> in one and "-p udp". Likewise you would need to update the graph URL 
>>>>>>> with whatever extra options it needs to grab the different results 
>>>>>>> (iirc, bandwidthBraph.cgi has a "protocol=udp/tcp" GET parameter you 
>>>>>>> could set in this example). I think another way to ask the question 
>>>>>>> is what types of test parameters do you want to distinguish between 
>>>>>>> tests and are they all supported as parameters to the graphs and 
>>>>>>> underlying tools? 
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Andy
>>>>>>>
>>>>>>>
>>>>>>> On Jun 11, 2014, at 9:22 AM, Chad Kotil 
>>>>>>> <>
>>>>>>>  wrote:
>>>>>>>
>>>>>>>> Hello perfsonar-users,
>>>>>>>> Is there a way to add multiple meshes to maddash with hosts that are 
>>>>>>>> a
>>>>>>>> member of multiple testspec's?
>>>>>>>>
>>>>>>>> I am trying to have maddash display multiple meshes which rely on
>>>>>>>> different testspec's as defined in my owmesh.conf. It doesnt look 
>>>>>>>> like
>>>>>>>> you can specify a testpec_id however in maddash right out of the box.
>>>>>>>> The old bwplot.cgi handles this case just fine , as it accepts a
>>>>>>>> testspec via the name parameter.
>>>>>>>>
>>>>>>>> I've done some digging and found that the service check command, the
>>>>>>>> graphUrl and metaDataKeyLookup in maddash.yaml do not seem to accept 
>>>>>>>> a
>>>>>>>> testspec_id. However I was able to modify the service check command,
>>>>>>>> check_throughput.pl to accept a testspec and modified the query to 
>>>>>>>> match
>>>>>>>> on the testspec, so that part does seem to be working. However 
>>>>>>>> without
>>>>>>>> metaKeyReq.cgi and delayGraph.cgi accepting a testspec_id I think I 
>>>>>>>> am
>>>>>>>> stuck.
>>>>>>>>
>>>>>>>> The only thing I can even think to do is use a unique MA, but that is
>>>>>>>> not ideal.
>>>>>>>>
>>>>>>>> Are there any other options available for what I am trying to do? And
>>>>>>>> are there any plans to add support for multiple meshes and testspecs
>>>>>>>> into maddash?
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> --Chad
>>>>>>>>
>>>>>>
>>>>
>>