[dpdk-dev] Number of memory channels (EAL Options)

[dpdk-dev] Number of memory channels (EAL Options)

[Muhammad Ali]

[-- Attachment #1: Type: text/plain, Size: 612 bytes --]

Hi everyone

I am running DPDK sample application compiled in Fedora 14 environment.

I have used following command to run the App

./<application> -c f -n 4

c= bitmask of cores to to be used.
and
n= number of memory channels

Now what does 'n' actually defines ?. Is this the number of DMA channels
between RAM (hugepages) and NIC. ? If not so then what it is ?

How does number of memory channels will effect the performance ?

BTW I am using hugepagesize=1G
and Number of Hugepages=4

My question might be very basic, but I am totally confused. Please help me
out here.

Thank you

Regards
-- 

Muhammad Ali

[-- Attachment #2: Type: text/html, Size: 1157 bytes --]

Re: [dpdk-dev] Number of memory channels (EAL Options)

[Stephen Hemminger]

On Thu, 20 Jun 2013 16:20:35 +0500
Muhammad Ali <Muhammad.Ali@xflowresearch.com> wrote:

> Hi everyone
> 
> I am running DPDK sample application compiled in Fedora 14 environment.
> 
> I have used following command to run the App
> 
> ./<application> -c f -n 4
> 
> c= bitmask of cores to to be used.
> and
> n= number of memory channels
> 
> Now what does 'n' actually defines ?. Is this the number of DMA channels
> between RAM (hugepages) and NIC. ? If not so then what it is ?
> 
> How does number of memory channels will effect the performance ?
> 
> BTW I am using hugepagesize=1G
> and Number of Hugepages=4
> 
> My question might be very basic, but I am totally confused. Please help me
> out here.
> 
> Thank you
> 
> Regards

Memory channels into the CPU, it controls the spread layout used by the memory
allocator, therefore it is a performance optimization.

Unfortunately, there is no good way to know what the number of memory channels
exactly without hardware specs. The BIOS may know but it is not exported in any
of the normal ACPI tables, and therefore not exposed by the OS.

What we end up doing in our startup script was
 "dmidecode -t 17 | grep -c 'Size:'" 
which will give you some idea of the number of populated memory banks and making
an educated guess on the number of channels. I.e 6 memory slots probably implies 3 channels.
If someone has a better way I would love to hear it.

Computing cpu bitmask isn't hard using "grep -c processor /proc/cpuinfo" and a
little maths in a shell script.

Re: [dpdk-dev] Number of memory channels (EAL Options)

[Muhammad Ali]

[-- Attachment #1: Type: text/plain, Size: 2683 bytes --]

Thanks for your detailed reply, I just want to clear myself on Number of
Memory Channels, These Memory channels are between

CPU and Hugepages (RAM)
or
HugePages (RAM) and NIC   (multi channel DMA)
or
both

?

Thanks


On Thu, Jun 20, 2013 at 8:43 PM, Stephen Hemminger <
stephen@networkplumber.org> wrote:

> On Thu, 20 Jun 2013 16:20:35 +0500
> Muhammad Ali <Muhammad.Ali@xflowresearch.com> wrote:
>
> > Hi everyone
> >
> > I am running DPDK sample application compiled in Fedora 14 environment.
> >
> > I have used following command to run the App
> >
> > ./<application> -c f -n 4
> >
> > c= bitmask of cores to to be used.
> > and
> > n= number of memory channels
> >
> > Now what does 'n' actually defines ?. Is this the number of DMA channels
> > between RAM (hugepages) and NIC. ? If not so then what it is ?
> >
> > How does number of memory channels will effect the performance ?
> >
> > BTW I am using hugepagesize=1G
> > and Number of Hugepages=4
> >
> > My question might be very basic, but I am totally confused. Please help
> me
> > out here.
> >
> > Thank you
> >
> > Regards
>
> Memory channels into the CPU, it controls the spread layout used by the
> memory
> allocator, therefore it is a performance optimization.
>
> Unfortunately, there is no good way to know what the number of memory
> channels
> exactly without hardware specs. The BIOS may know but it is not exported
> in any
> of the normal ACPI tables, and therefore not exposed by the OS.
>
> What we end up doing in our startup script was
>  "dmidecode -t 17 | grep -c 'Size:'"
> which will give you some idea of the number of populated memory banks and
> making
> an educated guess on the number of channels. I.e 6 memory slots probably
> implies 3 channels.
> If someone has a better way I would love to hear it.
>
> Computing cpu bitmask isn't hard using "grep -c processor /proc/cpuinfo"
> and a
> little maths in a shell script.
>
>


-- 


Muhammad Ali
Software Development Engineer
xFlow Research Inc. <http://www.xflowresearch.com/>
+92-323-965-4043 (GMT +5)
*Skype: *ali.wcct

This electronic message transmission contains information from the company
that may be proprietary, confidential and/or privileged.  The information
is intended only for the use of the individuals or entity named above.  If
you are not the intended recipient, be aware that any disclosure, copying
or distribution or use of the contents of this information is prohibited.
 If you have received this electronic transmission in error, please notify
the sender immediately by replying to the address listed in the “From:”
field.

[-- Attachment #2: Type: text/html, Size: 3664 bytes --]