astf limit option

[Andreas Bourges]

HI,

...I need some clarification regarding expected throughput using ASTF. Still using two XL710 cards (4 ports in total). The "limit" option is described as "limit : Limit the number of flows" - but it looks like this options is applied per core? So if I want to generated an elephant flow using line-rate, I can only use one core - and thus not reach 40g line rate? If I use more cores, I can reach line-rate, but on the other hand the flows are multiplied by the number of used cores and number of port-pairs (e.g. 12 cores, 2 port-pairs results in 24 flows).

Any hint?

Thanks,

Andreas

[hanoh haim]

H Andreas,

What a coincidence,I’ve just improved the eFlow performance, I could reach ~15Gb download per *one* flowI will push the example and code.

Regarding your question, it is not possible (in the current design)  to send the *same* flow from many cores. Flow is attached to core.

With limit=20 and threads=2 you will get total of 20 flows.

With limit=1 and threads=2 you will get 2 flows.

Thanks,

Hanoh

--
You received this message because you are subscribed to the Google Groups "TRex Traffic Generator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+u...@googlegroups.com.
To post to this group, send email to trex...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/e421e40d-7228-4456-96b7-10998eaf15d9%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Hanoh

Sent from my iPhone

[hanoh haim]

Hi,  The branch is under regression, 

this is the latest code 

https://github.com/hhaim/trex-core/tree/pipe_line

this is how I run it

$sudo ./t-rex-64 --astf -f astf/http_eflow2.py -m 10 -d 1000 -c 1 -l 1000 -t size=800,loop=100000,win=512,pipe=1 

this is from the profile 

# Example of elephant flow using loop of send (in the server side)

# 

#

#sudo ./t-rex-64 --astf -f astf/http_eflow2.py -m 10 -d 1000 -c 1 -l 1000 -t size=800,loop=100000,win=512,pipe=1

#

# size  : is in KB for each chuck in the loop (should be bigger than win for best performance)

# loops : How many chunks to download

# win   : in KB, the maximum window size. make it big for BDP  

# pipe  : Don't block on each send, 1 for maximum performance 

# mss   : the mss of the traffic 

#######

this is the results on my setup (with old 2Ghz machine)

each dual ports (c-s) has one flow.

thanks,

Hanoh

On Wed, May 2, 2018 at 4:34 PM, hanoh haim <hhaim...@gmail.com> wrote:

H Andreas,

What a coincidence,I’ve just improved the eFlow performance, I could reach ~15Gb download per *one* flowI will push the example and code.

Regarding your question, it is not possible (in the current design)  to send the *same* flow from many cores. Flow is attached to core.

With limit=20 and threads=2 you will get total of 20 flows.

With limit=1 and threads=2 you will get 2 flows.

Thanks,

Hanoh

On Wed, 2 May 2018 at 15:56 'Andreas Bourges' via TRex Traffic Generator <trex...@googlegroups.com> wrote:

HI,

...I need some clarification regarding expected throughput using ASTF. Still using two XL710 cards (4 ports in total). The "limit" option is described as "limit : Limit the number of flows" - but it looks like this options is applied per core? So if I want to generated an elephant flow using line-rate, I can only use one core - and thus not reach 40g line rate? If I use more cores, I can reach line-rate, but on the other hand the flows are multiplied by the number of used cores and number of port-pairs (e.g. 12 cores, 2 port-pairs results in 24 flows).

Any hint?

Thanks,

Andreas

--
You received this message because you are subscribed to the Google Groups "TRex Traffic Generator" group.

To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+unsubscribe@googlegroups.com.

To post to this group, send email to trex...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/e421e40d-7228-4456-96b7-10998eaf15d9%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Hanoh

Sent from my iPhone

[andyb...@googlemail.com]

Hi Hano,

 

…thanks a lot – currently downloading …

 

Regards,

 

Andreas

 

 

 

Von: hanoh haim <hhaim...@gmail.com>
Gesendet: Wednesday, May 2, 2018 15:44
An: Andreas Bourges <andyb...@googlemail.com>
Cc: TRex Traffic Generator <trex...@googlegroups.com>
Betreff: Re: [trex-tgn] astf limit option

 

Hi,  The branch is under regression, 

 

this is the latest code 

 

https://github.com/hhaim/trex-core/tree/pipe_line

 

 

this is how I run it

 

$sudo ./t-rex-64 --astf -f astf/http_eflow2.py -m 10 -d 1000 -c 1 -l 1000 -t size=800,loop=100000,win=512,pipe=1 

 

 

this is from the profile 

 

# Example of elephant flow using loop of send (in the server side)

# 

#

#sudo ./t-rex-64 --astf -f astf/http_eflow2.py -m 10 -d 1000 -c 1 -l 1000 -t size=800,loop=100000,win=512,pipe=1

#

# size  : is in KB for each chuck in the loop (should be bigger than win for best performance)

# loops : How many chunks to download

# win   : in KB, the maximum window size. make it big for BDP  

# pipe  : Don't block on each send, 1 for maximum performance 

# mss   : the mss of the traffic 

#######

 

 

this is the results on my setup (with old 2Ghz machine)

 

 

each dual ports (c-s) has one flow.

 

thanks,

Hanoh

 

 

On Wed, May 2, 2018 at 4:34 PM, hanoh haim <hhaim...@gmail.com> wrote:

H Andreas,

What a coincidence,I’ve just improved the eFlow performance, I could reach ~15Gb download per *one* flowI will push the example and code.

 

Regarding your question, it is not possible (in the current design)  to send the *same* flow from many cores. Flow is attached to core.

 

With limit=20 and threads=2 you will get total of 20 flows.

 

With limit=1 and threads=2 you will get 2 flows.

 

Thanks,

Hanoh

On Wed, 2 May 2018 at 15:56 'Andreas Bourges' via TRex Traffic Generator <trex...@googlegroups.com> wrote:

HI,

...I need some clarification regarding expected throughput using ASTF. Still using two XL710 cards (4 ports in total). The "limit" option is described as "limit : Limit the number of flows" - but it looks like this options is applied per core? So if I want to generated an elephant flow using line-rate, I can only use one core - and thus not reach 40g line rate? If I use more cores, I can reach line-rate, but on the other hand the flows are multiplied by the number of used cores and number of port-pairs (e.g. 12 cores, 2 port-pairs results in 24 flows).

Any hint?

Thanks,

Andreas

--
You received this message because you are subscribed to the Google Groups "TRex Traffic Generator" group.

To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+u...@googlegroups.com.

To post to this group, send email to

trex...@googlegroups.com.

To view this discussion on the web visit

https://groups.google.com/d/msgid/trex-tgn/e421e40d-7228-4456-96b7-10998eaf15d9%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Hanoh

Sent from my iPhone

[hanoh haim]

Hi, 

The same from XL710 setup with 4 ports (does not have LRO). 

flow_rate =15.2Gb/sec for one core/one flow. 

it would be interesting to see if Linux can do better than that with MSS=1460 

thanks,
Hanoh

To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+unsubscribe@googlegroups.com.

To post to this group, send email to

trex...@googlegroups.com.

To view this discussion on the web visit

https://groups.google.com/d/msgid/trex-tgn/e421e40d-7228-4456-96b7-10998eaf15d9%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Hanoh

Sent from my iPhone

 

--

Hanoh

Sent from my iPhone

[andyb...@googlemail.com]

Hi Hanoh,

 

…yes – I can confirm 15 Gbps on XL710. Changing MSS to 1460 does not change results (or did you mean standard linux kernel implementation).

 

Thanks and regards,

To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+u...@googlegroups.com.

To post to this group, send email to

trex...@googlegroups.com.

To view this discussion on the web visit

https://groups.google.com/d/msgid/trex-tgn/e421e40d-7228-4456-96b7-10998eaf15d9%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Hanoh

Sent from my iPhone

 

--

Hanoh

Sent from my iPhone

[hanoh haim]

I mean that usually high numbers reported with http download using Linux is done with MSS=9k that make it much easier to the DUT/Linux.

I did the test with default MSS (1460)

Thanks,

Hanoh