[Web10g-user] Fwd: Statistics validation

[Andrew K. Adams]

This thread seems to have been dropped from web10g-user — forwarded it (so we have record).


Andrew

Begin forwarded message:

> From: "Andrew K. Adams" <akadams at psc.edu>
> Subject: Re: [Web10g-user] Statistics validation
> Date: September 2, 2014 at 12:48:29 PM EDT
> To: hsirk <hsirk_6 at yahoo.com>
> Cc: "Andrew K. Adams" <akadams at psc.edu>, web10g-dev at web10g.org
> 
> Hi Krishnan, I apologize for responding so late to you on this!
> 
> I’ve updated DataOctetsOut in tcp_estats.c in both our (Web10g’s) mainline and 2.6 branch to be ‘end_seq - seq’; in my defense, ‘len’ in sk_buff is defined as "the data length”.  ;-)  I’ve additionally updated ThruOctetsReceived to be calculated (as you suggest) using the last window update, i.e., ‘tp->rcv_wup - seq’.  
> 
> Thanks for bringing these to our attention!
> 
> Out of curiosity, are you using our 2.6 branch?
> 
> 
> Andrew
> 
> --
> Andrew K. Adams				Senior Network Engineer
> Pittsburgh Supercomputing Center	Office:	361
> Carnegie Mellon University		Phone:	(412) 268-5142
> 300 South Craig Street			Fax:		(412) 268-5832
> Pittsburgh, PA 15213			WWW:	http://www.psc.edu/
> 
> 	D3 FA 7D 61 FD ED BD D9  0C DE 94 DB 0F 25 D0 2E
> 
> On Aug 6, 2014, at 8:26 PM, hsirk <hsirk_6 at yahoo.com> wrote:
> 
>> Resending ...bounced the first time.
>> 
>> On Wednesday, August 6, 2014 5:22 PM, hsirk <hsirk_6 at yahoo.com> wrote:
>> 
>> 
>> Hello,
>> I have been playing around with the web10g patch.
>> Back ported the web10g-v2.6.38 patch to centos6.5 (2.6.32-431).
>> Started validating the variables .  
>> 
>> So far i have queries about the below two stats.
>> 
>> 1.
>> tcp_estats_update_segsend()
>> stats->estats_vars.DataOctetsOut += len;
>> 
>> Based on the description of the variable,shouldn't the length be just the  
>> length of tcp data instead of the whole packet .(something line  seg - end_seg?)
>> VariableName:	DataOctetsOut
>> Scope:		RFC4898
>> ShortDescr:	Data Octets Out
>> Units:		octets
>> ProcType:	ZeroBasedCounter32
>> Description:	The number of octets of data contained in transmitted segments,
>> 		including retransmitted data.  Note that this does not include
>> 		TCP headers.
>> R/W:		read-only
>> 
>> 
>> 
>> 2.void tcp_estats_update_rcvd(struct tcp_sock *tp, u32 seq)
>>    stats->estats_vars.ThruOctetsReceived += seq - tp->rcv_nxt 
>> 
>> Based on the definition shouldn't rcv_wup be used ( ie something like tp->rcv_wup - tp->seq )
>> cause AFAIK rcv_nxt  indicates data that is received but not yet acked.
>> 
>> VariableName:
>> 	ThruOctetsReceived
>> Scope:		RFC4898
>> ShortDescr:	Total throughput octets received
>> Units:		octets
>> ProcType:	ZeroBasedCounter32
>> Description:	The number of octets for which cumulative acknowledgments
>> 		have been sent.  Note that this will be the sum of changes
>> 		to tcpEStatsAppRcvNxt.
>> R/W:		read-only
>> 
>> If my understanding is wrong or i am just looking at the wrong doc :). Please point me to the right one.
>> 
>> Regards
>> Krishnan
>> PS: if you would like the ported rh6.5 patch . Do let me know i will be more than happy to share it.