You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@trafficserver.apache.org by Emiliano Bonansea <em...@gmail.com> on 2014/08/22 16:09:32 UTC

IN-OUT traffic inconsistencies

Hi,

We're using ATS 5.0.1 in forward proxy mode and it has been working very 
well for nearly a month. To keep record of the traffic and calculate the 
savings we're measuring RX and TX bytes from this two linux counters.

/sys/class/net/eth0/statistics/rx_bytes
/sys/class/net/eth0/statistics/tx_bytes

The inconsistency that we see is that from time to time we can see that 
traffic IN is bigger that traffic OUT, wich is a bit odd since we double 
checked that there is no ATS external process consuming traffic, and 
disabled backgroud fetch plugin. Did anyone saw this behaviour before? 
from my perspective it is not normal. Is there any other method to 
measure the traffic without this inconsistency?
Below some examples using ifstat -b command.

        eth0
  Kbps in  Kbps out
23740.07  30806.81
*33503.20  33470.30*
*37809.30  35301.64*
34032.15  36976.01
*40354.97  40093.61*
*43426.62  40416.79*
38281.44  42624.68
33337.25  41107.99
31595.78  35656.48
35463.78  37135.69
26828.13  30668.24
*29982.00  28861.37*
28845.20  33033.17
*31570.18  29927.97**
**30895.75  33535.02**
**36162.10  33765.09*
36300.88  36656.24
37451.63  40036.47
*34232.43  33146.56**
**36534.28  36044.94*
34228.04  40689.40
*38079.69  35167.33*
36438.42  41970.43
39273.67  39591.09
38092.20  43111.67
33368.88  36600.54
*37749.38  35595.52*

Kind regards.
Emiliano
-- 
Interlink SRL

Re: IN-OUT traffic inconsistencies

Posted by "Alan M. Carroll" <am...@network-geographics.com>.
Emiliano,

We have had anecdotal reports of this but it has not been confirmed so there is not a lot of information. But it would be helpful to understand your network architecture a little better, in particular the relationship between ATS, eth0, the clients, and the origin servers. Most ATS deployments have 2 ethernet interfaces, one for clients and one for origin servers. Do you have only one? Or if you have two, which side is this interface on?

One thing to consider is that ATS by default doesn't meter its input side, that fetching from an origin server. that is, ATS will read data from an origin server as fast as it can, regardless of the client. If you have clients fetching large files, ATS can easily have more traffic from the origin servers than to the clients as it fills its internal buffers. This could, in theory, also cause a difference over the long term because if a client aborts a connection then you could see traffic for the full object on the input side but not on the output side.

Friday, August 22, 2014, 10:10:02 AM, you wrote:

> The inconsistency that we see is that from time to time we can see       that traffic IN is bigger that traffic OUT, wich is a bit odd       since we double checked that there is no ATS external process       consuming traffic, and disabled backgroud fetch plugin. Did anyone       saw this behaviour before? from my perspective it is not normal.       Is there any other method to measure the traffic without this       inconsistency?


Re: IN-OUT traffic inconsistencies

Posted by Emiliano Bonansea <em...@gmail.com>.
Hi,

I would like to make clear as well, that we're not using the read while 
write function or background fill.

regards.
Emiliano.
Interlink SRL

El 22/08/2014 11:09 a.m., Emiliano Bonansea escribió:
> Hi,
>
> We're using ATS 5.0.1 in forward proxy mode and it has been working 
> very well for nearly a month. To keep record of the traffic and 
> calculate the savings we're measuring RX and TX bytes from this two 
> linux counters.
>
> /sys/class/net/eth0/statistics/rx_bytes
> /sys/class/net/eth0/statistics/tx_bytes
>
> The inconsistency that we see is that from time to time we can see 
> that traffic IN is bigger that traffic OUT, wich is a bit odd since we 
> double checked that there is no ATS external process consuming 
> traffic, and disabled backgroud fetch plugin. Did anyone saw this 
> behaviour before? from my perspective it is not normal. Is there any 
> other method to measure the traffic without this inconsistency?
> Below some examples using ifstat -b command.
>
>        eth0
>  Kbps in  Kbps out
> 23740.07  30806.81
> *33503.20  33470.30*
> *37809.30  35301.64*
> 34032.15  36976.01
> *40354.97  40093.61*
> *43426.62  40416.79*
> 38281.44  42624.68
> 33337.25  41107.99
> 31595.78  35656.48
> 35463.78  37135.69
> 26828.13  30668.24
> *29982.00  28861.37*
> 28845.20  33033.17
> *31570.18  29927.97**
> **30895.75  33535.02**
> **36162.10  33765.09*
> 36300.88  36656.24
> 37451.63  40036.47
> *34232.43  33146.56**
> **36534.28  36044.94*
> 34228.04  40689.40
> *38079.69  35167.33*
> 36438.42  41970.43
> 39273.67  39591.09
> 38092.20  43111.67
> 33368.88  36600.54
> *37749.38  35595.52*
>
> Kind regards.
> Emiliano
> -- 
> Interlink SRL