Message-Id: <200109241708.f8OH8Qf29046@picard.cs.ohiou.edu> From: "Shawn Ostermann" <ostermann@cs.ohiou.edu> Subject: Re: Thruput versus sequence numbers Date: Mon, 24 Sep 2001 13:08:26 -0400
>> I was quite surprised when I saw my tcptrace thruput graph. It
>> indicates a thruput equals to 10e6 bytes per sec. The values of my
>> sequence numbers graph are totally different :
It's a little difficult to see the details from the pictures that you
sent. [Incidentally, Ryad had tried to send his question to the whole
list, but it bounced because it was over our message-size limit due to
the jpeg image].
If you want to see Ryad's jpeg file, look at:
http://irg.cs.ohiou.edu/private/sdo/ryad.jpg
You sent a jpeg showing a time sequence graph of a
fairly normal TCP conversation. It appears to have 6 congestion
events (from the red, retransmited segments). It appears to have sent
about 100,000 bytes over the span of of about 50 seconds. That works
out to ~80kbytes/second.
Measuring TCP throughput is difficult over short intervals. For
example, it's common for TCP to send back to back packets (during slow
start, for example). TCP would probably be able to send those at line
speed (the speed of the local net), but we clearly wouldn't consider
that to be the "throughput of TCP" at the time. Looking at the ACK
stream returning doesn't help, either, because ACKs can easily return
back-to-back as well, indicating an even higher "bytes acked"
throughput (even higher because the ACKs are smaller than the original
segment, so could be "closer together").
That's why tcptrace allows you to choose the interval over which you
want to calculate the throughput depending on what you want to know.
Sometimes knowing the line speed is helpful. I generally set the
interval to 16 segments as a first cut if I'm not sure what I'm
looking for exactly.
Your second JPEG appears to show the upper portion of the throughput
graph and shows a number of points in the 1MB/second range. I would
consider that possible over short runs of packets given the
explanation above. Of course, this is just speculation without seeing
the original trace file.
--sdo
-------------------------------------------------------------------------
Dr. Shawn Ostermann - Associate Professor - Ohio University
322B Stocker Center, Ohio University, Athens, Ohio 45701-2979
ostermann@cs.ohiou.edu -- FAX: (740)593-0007 -- Voice: (740)593-1234
http://ace.cs.ohiou.edu/~osterman http://irg.cs.ohiou.edu
>> Hie,
>>
>> I was quite surprised when I saw my tcptrace thruput graph. It
>> indicates a thruput equals to 10e6 bytes per sec. The values of my
>> sequence numbers graph are totally different :
>>
>> 10e6 per sec for thruput against 3*10e3 per seq for sequence number
>> !!!!!
>>
>> It's written on the website of Tcptrace, concerning thruput Graph, that
>> the points plotted are an average over 10 segments. Is it enough to
>> explain such a diffrence? Have I done a mistake?
>>
>> I've attached a jpg image for you to compare the measures.
>>
>> Thanks
>> Ryad
----------------------------------------------------------------------------
To unsubscribe, send a message with body containing "unsubscribe tcptrace" to
majordomo@tcptrace.org.
This archive was generated by hypermail 2b30 : 09/25/01 EDT
