NDT-DEV email list created

[Will Hawkins <>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Dr. Carlson,

Thank you for weighing in on this. This is exactly the type of thing
that we had to do to uncover why Flash speed throughput was
significantly slower than the results from the Java client on the same
machine. In fact, after looking at traces and TCP statistics it is
very obvious what the problem is.

The bad news was that the limitations were in the Flash runtime on
most platforms. The good news was that there were a few, select (but
common) combinations of OS and Flash runtime that did produce good
results.

After significant hand-wringing, we decided not to abandon a Flash
client altogether. We decided that an acceptable solution was to
release the Flash client but give the hoster/embedder the option to
warn or restrict the user when his/her Flash runtime was going to
generate bad results.

As I said in my previous message, I am going to try to dig back
through my notes and send around all the supporting evidence for our
conclusion. Not that it's anything special but because I thought it
was incredibly interesting.

I hope that all this makes sense. Given what I've said, I'd love to
hear any additional feedback or input you can offer. You are the
expert on this type of thing and Jordan and I were/are merely
apprentices learning from you, Aaron, MM, the code and others.

Thanks again for your feedback!
Will

On 04/06/2015 09:13 PM, Richard Carlson wrote:
> Aaron et.al.
> 
> The whole point of NDT is to gather the kernel data to answer
> questions like this.  So, what did the kernel collect?  Look at the
> server's log file and see what it says.  Tell the server to run
> with the ndttrace and snaplog options turned on.  You'll get a
> packet level trace you can run through something like wireshark and
> a 5msec report of every Web100 variable for the 10 sec S2C test.
> 
> You don't need to guess why the performance sucks, just ask TCP
> what the hell it was doing!
> 
> Rich
> 
> PS: this is exactly why just reporting throughput speeds is the
> wrong thing to do.  You want to report WHY the test spit out those
> numbers. The Web100 kernel gives you some glimpse into that
> process.  Use it!
> 
> On 04/06/2015 01:58 PM, Aaron Brown wrote:
>> Does anyone have a good explanation for these horrible
>> performance results? Is it really just that flash is
>> unpredictably awful on every platform but windows?
>> 
>> Cheers, Aaron
>> 
>>> Begin forwarded message:
>>> 
>>> *Date: *April 6, 2015 at 11:48:09 AM EDT *From: *Jeremy Palmer
>>> <
>>>  
>>> <mailto:>>
>>>  
>>> *To: *Kenneth D Miller II 
>>> <
>>> <mailto:>>,
>>>  Aaron Brown 
>>> <
>>> <mailto:>>
>>>  *Cc: 
>>> *<
>>> <mailto:>>
>>>  *Subject: **Re: Final(?)
>>> Release Candidate for NDT 3.7.0*
>>> 
>>> I see the same issue with lower speeds when using Flash
>>> compared to Java when running to your test host. In my case
>>> tho, I was seeing much lower upload speeds with Flash compared
>>> to Java. Here were my results (Windows 7, Java 8u31, Flash
>>> 17.0.0.134, FFox 37.0).
>>> 
>>> Flash: 1.7Mb/s UP, 29.5Mb/s DOWN Java: 10.1Mb/s UP, 29.7Mb/s
>>> DOWN
>>> 
>>> I ran multiple tests and each time results were similar to the
>>> above. Jeremy Palmer Senior Network Engineer ViaWest, Inc. 
>>> Office: 720.891.1045 Fax: 303-874-5236 http://www.viawest.com 
>>> On 4/6/2015 9:29 AM, Kenneth D Miller II wrote:
>>>> Hi Aaron,
>>>> 
>>>> Comments: - It seems like the flash version in Safari and
>>>> Firefox is limited in OSX.  Maybe its me, you can test my
>>>> server from http://128.118.46.248:7123/.  I just did a rpm
>>>> -Uvh to the 2 URLs below and rebooted this test server. -
>>>> Chrome threw a warning about Flash and Upload was less the
>>>> Safari and Firefox
>>>> 
>>>> Here is some testing I did.
>>>> 
>>>> From OSX: Safari Java: Up: 903.31mb/s Down: 941.33mb/s
>>>> 
>>>> Safari Flash: Up: 832.42mb/s Down: 30.51mb/s
>>>> 
>>>> 
>>>> Firefox Java: Up: 877.66 mb/s Down: 941.41mb/s
>>>> 
>>>> Firefox Flash: Up: 809.62mb/s Down: 30.55mb/s
>>>> 
>>>> 
>>>> Chrome Java: Up: 721.76 mb/s 941.38 mb/s
>>>> 
>>>> Chrome Flash: Warning: Using the flash client in Chrome may
>>>> provide less accurate results Up: 499.03 mb/s Down: 919.25
>>>> mb/s
>>>> 
>>>> 
>>>> I have been downloading the ndt client for some time onto my
>>>> laptop and run it from the CLI.  I scp /usr/ndt/Tcpbw100.jar
>>>> to a directory, create a sub-directory called lib and
>>>> download json-simple-1.1.1.jar. Then I run sudo java -jar
>>>> Tcpbw100.jar t248.npidev.psu.edu <http://t248.npidev.psu.edu>
>>>> 3001 and get the following java GUI results:
>>>> 
>>>> Network Diagnostic Tool Client v3.7.0-rc1 Click START to
>>>> start the test ** Starting test 1 of 1 ** Connecting to
>>>> 't248.npidev.psu.edu <http://t248.npidev.psu.edu>' 
>>>> [t248.npidev.psu.edu/128.118.46.248 
>>>> <http://t248.npidev.psu.edu/128.118.46.248>] to run test 
>>>> Connected to: t248.npidev.psu.edu
>>>> <http://t248.npidev.psu.edu>-- Using IPv4 address WARNING:
>>>> NDT server has different version number (3.7.0-rc3) Checking
>>>> for Middleboxes . . . . . . . . . . . . . . . . . . Done. 
>>>> Checking for firewalls . . . . . . . . . . . . . . . . . . .
>>>> Done. running 10s outbound test (client-to-server [C2S]) . .
>>>> . . . 858.44Mb/s running 10s inbound test (server-to-client
>>>> [S2C]) . . . . . . 941.33Mb/s Sending META information . . .
>>>> . . . . . . . . . . . . . . . . Done. The slowest link in the
>>>> end-to-end path is a 1.0 Gbps Gigabit Ethernet subnet Click
>>>> START to re-test
>>>> 
>>>> 
>>>> Here are the CLI results and output: 
>>>> DHCP-128-118-46-198:Downloads kdm193$ sudo java -jar
>>>> Tcpbw100.jar t248.npidev.psu.edu <http://t248.npidev.psu.edu>
>>>> 3001 Password: wait flag received = 0 Server version:
>>>> 3.7.0-rc3-Web100 WARNING: NDT server has different version
>>>> number (3.7.0-rc3-Web100) 57778772 bytes 84026.57262315943
>>>> kb/s 5.501 secs Sending '84026.57262315943' back to server 
>>>> calling in2Socket.getLocalAddress() SFW: port=42563 SFW:
>>>> testTime=1 SFW: oport=63435 java.net.ConnectException:
>>>> Connection refused at
>>>> java.net.PlainSocketImpl.socketConnect(Native Method) at 
>>>> java.net.AbstractPlainSocketImpl.doConnect(AbstractPlainSocketImpl.
java:345)
>>>>
>>>>
>>>> 
at
>>>> java.net.AbstractPlainSocketImpl.connectToAddress(AbstractPlainSock
etImpl.java:206)
>>>>
>>>>
>>>> 
at
>>>> java.net.AbstractPlainSocketImpl.connect(AbstractPlainSocketImpl.ja
va:188)
>>>>
>>>>
>>>> 
at java.net.SocksSocketImpl.connect(SocksSocketImpl.java:392)
>>>> Simple firewall test: Improper message at
>>>> java.net.Socket.connect(Socket.java:589) at
>>>> edu.internet2.ndt.Tcpbw100.test_sfw(Tcpbw100.java:1653) at
>>>> edu.internet2.ndt.Tcpbw100.dottcp(Tcpbw100.java:2801) at
>>>> edu.internet2.ndt.Tcpbw100$TestWorker.run(Tcpbw100.java:615) 
>>>> at java.lang.Thread.run(Thread.java:745) ******Send buffer
>>>> size =8192 SocketException while writing to
>>>> serverjava.net.SocketException: Socket closed 10005.0
>>>> millisec test completed,65536,16414 860136.2550724638 kb/s
>>>> outbound 1178075136 bytes 941330.5121853775 kb/s 10.012 secs 
>>>> Sending '941330.5121853775' back to server USERAGENT null 
>>>> Calling InetAddress.getLocalHost() twice
>>>> 
>>>> 
>>>> 
>>>> 
>>>> Thanks, Ken
>>>> 
>>>> Network Systems Engineer Network Development Engineering,
>>>> TNS, ITS Penn State University My Calendar Free/Busy - 
>>>> https://ucs.psu.edu/home//Calendar.html
>>>>  Chair
>>>> Internet2 Performance Working Group SuperComputing SCinet
>>>> Measurement Team Member
>>>> 
>>>> 
>>>> 
>>>> 
>>>> 
>>>> 
>>>> -------------------------------------------------------------------
- -----
>>>>
>>>>
>>>>
>>>> 
*From: *"Aaron Brown" 
<>
>>>> *To: 
>>>> *
>>>>  *Sent: *Friday, April 3, 2015
>>>> 3:07:23 PM *Subject: *Final(?) Release Candidate for NDT
>>>> 3.7.0
>>>> 
>>>> Hi All,
>>>> 
>>>> We have the, hopefully, final release candidate for a new
>>>> NDT release. This release candidate includes a number of
>>>> fixes and changes since -rc1.
>>>> 
>>>> Among other things, version 3.7.0 will include: - A java
>>>> client that works with Java 8 - A flash-based client - A new
>>>> web GUI - Support for using Apache instead of ‘fakewww'
>>>> 
>>>> This is a beta-test version, and is not recommended for
>>>> general use yet. However, we're hoping that, with some
>>>> testing in the community, this can be improved quickly and a
>>>> production release can happen relatively soon. We're
>>>> especially interested in feedback on how easy it is to use
>>>> the various new tools, and features.
>>>> 
>>>> Tarball:
>>>> 
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-3.7.0-r
c3.tar.gz
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-3.7.
0-rc3.tar.gz>
>>>>
>>>>
>>>>
>>>>
>>>> 
RH6/x86:
>>>> 
>>>> 
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-client-
3.7.0-4.el6.i686.rpm
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-clie
nt-3.7.0-4.el6.i686.rpm>
>>>>
>>>>
>>>>  
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-server-
3.7.0-4.el6.i686.rpm
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-serv
er-3.7.0-4.el6.i686.rpm>
>>>>
>>>>
>>>>  
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-server-
apache-3.7.0-4.el6.i686.rpm
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-serv
er-apache-3.7.0-4.el6.i686.rpm>
>>>>
>>>> 
(only
>>>> needed to use Apache instead of fakewww)
>>>> 
>>>> RH6/x86-64:
>>>> 
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-client-
3.7.0-4.el6.x86_64.rpm
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-clie
nt-3.7.0-4.el6.x86_64.rpm>
>>>>
>>>>
>>>>  
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-server-
3.7.0-4.el6.x86_64.rpm
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-serv
er-3.7.0-4.el6.x86_64.rpm>
>>>>
>>>>
>>>>  
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-server-
apache-3.7.0-4.el6.x86_64.rpm
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-serv
er-apache-3.7.0-4.el6.x86_64.rpm>
>>>>
>>>> 
(only
>>>> needed to use Apache instead of fakewww)
>>>> 
>>>> SRPM:
>>>> 
>>>> http://ndb1.internet2.edu/~aaron/ndt-3.7.0-4.el6.x86_64/ndt-3.7.0-4
.el6.src.rpm
>>>>
>>>>
>>>>  
>>>> <http://ndb1.internet2.edu/%7Eaaron/ndt-3.7.0-4.el6.x86_64/ndt-3.7.
0-4.el6.src.rpm>
>>>>
>>>>
>>>>
>>>>
>>>> 
Cheers,
>>>> Aaron
>>>> 
>>>> 
>>> 
>> 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
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=d72M
-----END PGP SIGNATURE-----