[Users] Torus2q routing engine and vlarb weight choise

[Hal Rosenstock]

Please read doc/qos-config.txt and IBA spec vol 1 section 7.6.9.2 VL
ARBITRATION
WHEN MULTIPLE DATA VL S ARE IMPLEMENTED

It's max_vls parameter which relates to VLs (and hence buffers).

These parameters are for the VL arbitration. A weight of 0 means skip entry.

Since t2q assigns qos-level to group of 4 VLs, I think weighting should be
same on those VLs.

I also would try not to have some VL hog the arbitration so I'd use small
weights.

So something like:

qos_ca_max_vls 1
qos_ca_vlarb_high 0:4,1:4,2:0,3:0,4:0,5:0,6:0,7:0
qos_ca_vlarb_low  0:4,1:4,2:0,3:0,4:0,5:0,6:0,7:0

qos_swe_max_vls 4
qos_swe_vlarb_high* 0:4,1:4,2:4,3:4*,4:0,5:0,6:0,7:0
qos_swe_vlarb_low  *0:4,1:4,2:4,3:4*,4:0,5:0,6:0,7:0

if you want to use both high and low priority VL Arbitration but you could
probably also only use the low or high priority VL arbitration too. I
haven't played with all the variations for this but the general principles
of how t2q uses QoS are expressed above.





On Fri, Aug 5, 2016 at 9:43 AM, Black.S <Black.S52 at yandex.com> wrote:

> Thanks for reply
> I use only one QoS level.
> Correct me please.
>
> 1) Is that config is right for 3D torus ?
>
>     qos_ca_vlarb_high 0:4,1:0,2:0,3:0,4:0,5:0,6:0,7:0
>     qos_ca_vlarb_low  0:4,1:4,2:0,3:0,4:0,5:0,6:0,7:0
>
>     qos_swe_vlarb_high 0:4,1:0,2:0,3:0*,*4:0,5:0,6:0,7:0     # because
> Torus use VL0-3
>     qos_swe_vlarb_low  0:4,1:4,2:0,3:0,4:0,5:0,6:0,7:0
>
> or this right  ?
>     qos_vlarb_high *0:4,1:4,2:4,3:4*,4:0,5:0,6:0,7:0
>     qos_vlarb_low  *0:4,1:4,2:4,3:4*,4:0,5:0,6:0,7:0
>
>     qos_swe_vlarb_high* 0:4,1:4,2:4,3:4*,4:0,5:0,6:0,7:0     # because
> Torus use VL0-3
>     qos_swe_vlarb_low  *0:4,1:4,2:4,3:4*,4:0,5:0,6:0,7:0 2)
>
>
> 2) And still Which weight prefer for VL0-3 ? Why not prefer weight 255 ?
>
> Or perhaps that I do not understand ?
>
> Thanks all
>
>
> 04.08.2016 20:26, Hal Rosenstock пишет:
>
> On links between end ports and peer switch, only 2 VLs max are needed
> (depending on whether 1 or 2 QoS levels are in use). Interswitch links
> require 4 VLs per QoS level.
>
> With the config you are using, t2q is only warning that resources
> (buffers) are being wasted on unused VLs. It's up to you whether to do more
> config of QoS to get better buffer use and eliminate this warning message.
>
> On Thu, Aug 4, 2016 at 12:09 PM, B S <black.s52 at yandex.com> wrote:
>
>> Hello all
>> I can't understand some torus2q options.May be anybody explain me ?
>> In the Internet I found some options for torus
>>
>> qos_max_vls 8       # its clear for use 8 vl
>> qos_high_limit 0.   # its cleasr for processing only 1 message in high
>> table
>> qos_vlarb_low 0:4,1:4,2:0,3:0,5:0,6:0,7:0
>> qos_vlarb_high 0:4,1:0,2:0,3:0,4:0,5:0,6:0,7:0
>>
>> If I leave this options by default, opensm get warning
>> "Warning: torus-2QoS recommends 0 VLarb weights
>> for VLs 2-7 on endpoint links; not true ..."
>> Sorry for more questions. I can't found answer in internet.
>>
>> 1) Why I can't leave this tables ( vlarb_high, vlarb_low) by default ?
>> Who in routing engine torus2q and how  use this table ?
>> 2) Why used only vl0 and vl1. In manual about Torus2q use 0-3 and 4-7 VLs
>> ?
>> 3) I can't understand why used weight 4 but not for example 64 or 255 ?
>> 4) Why I should write this table for extended ca, swe, etc. separately
>>
>> May be somebody have links about this theme ?
>> Thanks for you interest and time
>>
>> _______________________________________________
>> Users mailing list
>> Users at lists.openfabrics.org
>> http://lists.openfabrics.org/mailman/listinfo/users
>>
>>
>
>
> _______________________________________________
> Users mailing list
> Users at lists.openfabrics.org
> http://lists.openfabrics.org/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/users/attachments/20160805/9b035c0d/attachment.html>