ndt-users list created

[Richard Carlson <>]

Joe;

One minor drawback to TCP is that the sender knows quite a bit about 
what is going on, but the receiver doesn't.  This means that a ton of 
web100 data can be collected during the S2C test, but very little is 
obtained during the C2S test.  So the numbers you give below don't 
really tell us a lot.  There is a "-r" flag for the NDT testing 
engine (web100srv) that will record the C2S statistics, but it is 
turned off by default.  See the man page for more details.  You can 
add it in and maybe that will tell you something.

A couple of thoughts about the data below.

Both WinScaleSent and WinScaleRcvd are -1.  For reasons known to the 
Linux developers, Linux doubles the buffer size so 128KB is really a 
64 KB Window and window scaling is off.  This could be part of your problem.

Have you done any TCP tuning on the server?  Look at the NDT INSTALL 
file.  Step 9.d. gives you some server tuning guide lines.

What is the client and what else is it doing?  If possible look at 
the client's CPU utilization while the test is running.  It may be a 
CPU limit instead of a network infrastructure limit.

Regards;
Rich

At 01:17 PM 8/24/2007, Joe Freeman wrote:
> Greetings-
>
> I have NDT v5.4.12 (as reported in the applet screen) setup on an 
> ubuntu server (2.6.21-web100). S2C appears to run fine, but but my 
> numbers are way low on the C2S side when I test with high speeds.
>
> From one test point (100M connection, 4ms rtt), I only get 19M C2S.
>
> Here's the info I have from the web100 stats...
>
> WEB100 Kernel Variables
> Client: localhost/127.0.0.1
> CurMSS: 1260
> X_Rcvbuf: 262142
> X_Sndbuf: 262142
> AckPktsIn: 25976
> AckPktsOut: 0
> BytesRetrans: 6300
> CongAvoid: 0
> CongestionOverCount: 0
> CongestionSignals: 1
> CountRTT: 25937
> CurCwnd: 100800
> CurRTO: 211
> CurRwinRcvd: 65520
> CurRwinSent: 5840
> CurSsthresh: 79380
> DSACKDups: 0
> DataBytesIn: 0
> DataBytesOut: 58701500
> DataPktsIn: 0
> DataPktsOut: 46221
> DupAcksIn: 32
> ECNEnabled: 0
> FastRetran: 1
> MaxCwnd: 100800
> MaxMSS: 1260
> MaxRTO: 220
> MaxRTT: 113
> MaxRwinRcvd: 65520
> MaxRwinSent: 5840
> MaxSsthresh: 126000
> MinMSS: 1260
> MinRTO: 205
> MinRTT: 4
> MinRwinRcvd: 0
> MinRwinSent: 5840
> NagleEnabled: 1
> OtherReductions: 1
> PktsIn: 25976
> PktsOut: 46221
> PktsRetrans: 5
> RcvWinScale: -1
> SACKEnabled: 3
> SACKsRcvd: 13
> SendStall: 0
> SlowStart: 0
> SampleRTT: 11
> SmoothedRTT: 11
> SndWinScale: -1
> SndLimTimeRwin: 9936256
> SndLimTimeCwnd: 80051
> SndLimTimeSender: 59056
> SndLimTransRwin: 4
> SndLimTransCwnd: 4
> SndLimTransSender: 1
> SndLimBytesRwin: 58357780
> SndLimBytesCwnd: 343720
> SndLimBytesSender: 0
> SubsequentTimeouts: 0
> SumRTT: 288976
> Timeouts: 0
> TimestampsEnabled: 0
> WinScaleRcvd: -1
> WinScaleSent: -1
> DupAcksOut: 0
> StartTimeUsec: 416784
> Duration: 10075375
> c2sData: 4
> c2sAck: 4
> s2cData: 9
> s2cAck: 9
> half_duplex: 0
> link: 100
> congestion: 0
> bad_cable: 0
> mismatch: 0
> spd: 46.61
> bw: 185.50
> loss: 0.000021635
> avgrtt: 11.14
> waitsec: 0.00
> timesec: 10.00
> order: 0.0012
> rwintime: 0.9862
> sendtime: 0.0059
> cwndtime: 0.0079
> rwin: 0.4999
> swin: 2.0000
> cwin: 0.7690
> rttsec: 0.011141
> Sndbuf: 262142
> aspd: 0.00000
> CWND-Limited: 1506.36
>
> The theoretical network limit is 185.5 Mbps
> The NDT server has a 127.0 KByte buffer which limits the throughput 
> to 179.51 Mbps
> Your PC/Workstation has a 63.0 KByte buffer which limits the 
> throughput to 44.87 Mbps
> The network based flow control limits the throughput to 69.02 Mbps
>  Client Data reports link is 'T3', Client Acks report link is 'T3'
> Server Data reports link is '10 Gig', Server Acks report link is '10 Gig'
>
> The server link is actually 100M Ethernet for the moment it'll go to 
> 1G when I get the correct sfp for my router.
>
> One thing I can't figure out is that NDT always seems to report a 
> server buffer of 127KB, even on much slower connections. If 
> autotuning is used (it is enabled.. I double checked), I'd expect 
> that number to change as well.. at least in variation with the rtt 
> between the client and server.
>
> Any thoughts as to where I should be looking?
>
> Thanks-
> Joe

------------------------------------



Richard A. Carlson                              e-mail: 

Network Engineer                                phone:  (734) 352-7043
Internet2                                       fax:    (734) 913-4255
1000 Oakbrook Dr; Suite 300
Ann Arbor,  MI  48104