But, flit latency keeps growing unbounded as I increase the injection rate (as 
if buffer was infinite).
Good question. The buffers are finite everywhere except at the traffic 
generator (src/cpu/testers/GarnetSyntheticTraffic/).
The traffic generator keeps creating new packets at the user-specified 
injection rate. But the rate of their injection into the network of course gets 
saturated. So the latency shoots up because the “queueing latency” factor keeps 
growing while the “network latency” component saturates. You can see this 
phenomenon in the stats if you look at the queueing and network latency 
breakdowns.

The saturation throughput is the injection rate at which the latency shoots up.
Alternately, you can plot the total packets received per cycle as a function of 
injection rate, and you will see that number saturating.

Cheers,
Tushar

On Jul 3, 2019, 11:08 AM -0400, CARLOS ANDRES PIEDRAHITA VELASQUEZ 
<[email protected]>, wrote:
Thank you Kushar for your kind response, now I get it.

I have one more question though, in SwitchAllocator.cc in the method 
send_allowed, for flits that are HEAD or HEADTAIL, why is not necessary credit 
checking? I'm asking this because I'm trying to reach saturation throughput in 
a Mesh 4x4 injecting flits in vnet 0 (one-flit vnet with 1 flit buffer depth by 
default). But, flit latency keeps growing unbounded as I increase the injection 
rate (as if buffer was infinite).

Thank for your attention.

On Mon, Jul 1, 2019 at 1:19 AM Krishna, Tushar 
<[email protected]<mailto:[email protected]>> wrote:
Those values are used when initializing the credits in OutVCstate for the data 
and control VCs.
The flitBuffer object has infinite size but the actual number of flits that can 
sit in it are controlled by the credits.

- Tushar
On Jun 27, 2019, 2:44 PM -0700, CARLOS ANDRES PIEDRAHITA VELASQUEZ 
<[email protected]<mailto:[email protected]>>, wrote:
Greetings Everyone,

I want to modify the buffer depth of virtual channels in Garnet from the 
default values (buffers_per_ctrl_vc = 1 and buffers_per_data_vc = 4). Analyzing 
GarnetNetwork.cc, I can see that these attributes are instantiated in 
GarnetNetwork's constructor (lines 65 and 66), but never used anywhere in the 
code. Additionally, In VirtualChannel.cc  the attribute m_input_buffer is 
instantiated as a flitBuffer object with infinite size (line 40).

So, does this mean that the buffers in the virtual channels are of infinite 
size?, is there something that I don't understand yet?

Thank you for your kind attention.

--
Carlos A. Piedrahita-Velásquez
Universidad de Antioquia

[https://0c595a64-a-57bd039f-s-sites.googlegroups.com/a/udea.edu.co/repositorio-logos/home/firma/FIRMA_CAMPUS50.jpg]
"La información aquí contenida es para uso exclusivo de la persona o entidad de 
destino. Está estrictamente prohibida su utilización, copia, descarga, 
distribución, modificación y/o reproducción total o parcial, sin el permiso 
expreso de Universidad de Antioquia, pues su contenido puede ser de carácter 
confidencial y/o contener material privilegiado. Si usted recibió esta 
información por error, por favor contacte en forma inmediata a quien la envió y 
borre este material de su computador. Universidad de Antioquia no es 
responsable por la información contenida en esta comunicación, el directo 
responsable es quien la firma o el autor de la misma."

_______________________________________________
gem5-users mailing list
[email protected]<mailto:[email protected]>
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users


--
Carlos A. Piedrahita-Velásquez
Universidad de Antioquia
Medellín - Colombia

[https://0c595a64-a-57bd039f-s-sites.googlegroups.com/a/udea.edu.co/repositorio-logos/home/firma/FIRMA_CAMPUS50.jpg]
"La información aquí contenida es para uso exclusivo de la persona o entidad de 
destino. Está estrictamente prohibida su utilización, copia, descarga, 
distribución, modificación y/o reproducción total o parcial, sin el permiso 
expreso de Universidad de Antioquia, pues su contenido puede ser de carácter 
confidencial y/o contener material privilegiado. Si usted recibió esta 
información por error, por favor contacte en forma inmediata a quien la envió y 
borre este material de su computador. Universidad de Antioquia no es 
responsable por la información contenida en esta comunicación, el directo 
responsable es quien la firma o el autor de la misma."

_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to