RE: Wierd metro WiFi problem

I would suggest pinging with larger sized icmp packets (1000 bytes or
so).  While you may not be seeing much issue transmitting 64 bytes, your
chances of collision and data loss go up many, many times with a ~1000
byte frame.



> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On Behalf Of 
> [email protected]
> Sent: Thursday, January 26, 2006 3:08 PM
> To: SeattleWireless Talk List
> Subject: Re: Wierd metro WiFi problem
> 
> Tyler,
> 
> Do the drops vary with packet size? We've had cases of 
> equipment interfaces being good enough for pings and 
> traceroutes but just flaky enough to truncate larger packets.
> 
> dgl
> 
> On Thursday, January 26, 2006, at 02:48PM, Tyler van 
> Houwelingen <[email protected]> wrote:
> 
> >My netstat -s is below.  This is across 2 5GHz wireless hops 
> (1 mile each), 
> >one of which has about 2Mbps of constant metro WiFi traffic. 
>   I have about 
> >10ms of latency back to the Main Internet router from my 
> test.  Links 
> >quality is exceptional on both.
> >
> >Recieved packets discarded:  234
> >Recieved packets delivered:   4180
> >TCP Segments recieved:  2583
> >TCP Segments Sent:  2271
> >TCP Segments retransmitted:  15
> >
> >Can anyone recomend a nice tool or utility to watch the 
> receive errors? 
> >They seem to be getting out of order or some other issue.
> >
> >Tyler
> >
> >
> >----- Original Message -----
> >From: "Alberto Escudero-Pascual" <[email protected]>
> >To: "Tyler van Houwelingen" <[email protected]>
> >Cc: "Claude" <[email protected]>; "Alberto Escudero-Pascual" 
> ><[email protected]>; "SeattleWireless Talk List" 
> ><[email protected]>; "Juan Rivero" <[email protected]>; 
> >"Dave Tapscott" <[email protected]>; "Paul Stark" 
> <[email protected]>; "Greg Vert"
> ><[email protected]>; "Marc Van Houwelingen" 
> ><[email protected]>; "Jose Manuel Gonzalez Saez" 
> ><[email protected]>; <[email protected]>
> >Sent: Thursday, January 26, 2006 1:09 PM
> >Subject: Re: Wierd metro WiFi problem
> >
> >
> >> Hi Tyler,
> >>
> >> What do you mean by "you add a little latency" to the 
> download. Do you
> >> artificially add delay into one of the hops? How is your testing PC
> >> connected to the backbone?
> >>
> >> The best way to ensure that it is not a TCP implementation related
> >> behavior is to run some kind of UDP stateless tests.
> >>
> >> Check the following link (Section 2.1), might give you some hints!
> >>
> >> http://www.29west.com/docs/THPM/tcp-latency.html
> >>
> >> Alberto
> >>
> >>
> >>
> >> On Thu, 2006-01-26 at 12:05 -0700, Tyler van Houwelingen wrote:
> >>> I am contacting the best wireless minds I know to help 
> with a bizarre
> >>> issue that has us stumped.
> >>>
> >>> Situation:
> >>> - We have 4 bonded T1 lines currently for our backbone, running a
> >>> large (400 AP) metro WiFi network.
> >>> - When connecting directly into the backbone router, we 
> get 6M down,
> >>> 6M up as expected.
> >>> - However, when we begin to add a little latency between 
> a testing PC
> >>> and the backbone connection, the upload speed stays the 
> same, but the
> >>> download gets hammered.  With 10ms of average latency 
> from the metro
> >>> WiFi network side, the speed drops to 2M down and 5M up.  
> At 40ms, it
> >>> is around 800k/5M.
> >>> - The wireless communications are more than fine as we 
> get 8M/8M back
> >>> to our headquarters from anywhere in the metro WiFi network.  The
> >>> problems only come from testing into the Internet.  We 
> get the same
> >>> goofy results when we disconnect the city network 
> completely and just
> >>> add latency in the lab.
> >>> - We have run all the DrTCP (MTU, RWIN) like tricks which provides
> >>> some improvement, but not much.
> >>>
> >>> Why would this additional latency kill the downloads so much?  I
> >>> suspect the problem is with our backbone provider 
> (Expedius) or some
> >>> limitation of metro WiFi that worries me.  In our other metro WiFi
> >>> network, we do not see this, but they all use a different backbone
> >>> provider (AT&T).
> >>>
> >>> Any suggestions to debug/fix this would be very much appreciated.
> >>>
> >>> tyler
> >>>
> >>>
> >>>
> >>> Tyler van Houwelingen
> >>> Founder and CEO
> >>> Azulstar Networks, Inc.
> >>> 1051 Jackson, Grand Haven, MI 49417
> >>> 1117 Rio Rancho Blvd 13, Rio Rancho, NM 87124
> >>> 1-877-AZULSTAR
> >>> www.azulstar.com
> >>
> >> 
> >
> >
> >_______________________________________________
> >Talk mailing list
> >[email protected]
> >http://seattlewireless.net/mailman/listinfo/talk
> >
> >
> _______________________________________________
> Talk mailing list
> [email protected]
> http://seattlewireless.net/mailman/listinfo/talk
> 
_______________________________________________
Talk mailing list
[email protected]
http://seattlewireless.net/mailman/listinfo/talk