Josh,
Note this test is doing two things:
PULL DMA where the FPGA is reading from system memory, with various read
latencies as Shep says, plus PUSH DMA where the FPGA is writing (back)
to system memory.
If the FPGA is pushing to another FPGA, or if the only stream is
FPGA->system memory, the numbers would also be higher.
We have a "pattern generator" worker that can be used to construct a
push-only FPGA->system test.
When you say "framework performance testing", can you suggest a few
things you are thinking of measuring?
There are a variety of capabilities for instrumenting and measuring
performance....
Cheers,
Jim
On 8/24/12 6:07 AM, Shepard Siegel wrote:
Hi Josh,
It's rough around the dges, but the script testBWcommand in my
upstream bin directory sweeps the PCIe message size and runs a number
of push or pull DMAs to report data like
Bytes 16000, Time delta = 1947, 8.217771 MBytes/seconds, Framesize 16
Bytes 32000, Time delta = 1960, 16.326531 MBytes/seconds, Framesize 32
Bytes 64000, Time delta = 1981, 32.306916 MBytes/seconds, Framesize 64
Bytes 128000, Time delta = 2040, 62.745098 MBytes/seconds, Framesize 128
Bytes 256000, Time delta = 2140, 119.626168 MBytes/seconds, Framesize 256
Bytes 512000, Time delta = 2473, 207.035989 MBytes/seconds, Framesize 512
Bytes 1024000, Time delta = 2862, 357.791754 MBytes/seconds, Framesize
1024
Bytes 2048000, Time delta = 3242, 631.708822 MBytes/seconds, Framesize
2048
Bytes 4096000, Time delta = 4832, 847.682119 MBytes/seconds, Framesize
4096
Bytes 8192000, Time delta = 8860, 924.604966 MBytes/seconds, Framesize
8192
These numbers, about half what you would expect with Gen1 x8 or Gen2
x4 are largely "held back" by the fact that our "pull" DMA engine has
only a single outstanding read. With two or more outstanding reads;
the pull DMA would see higher throughout with small message sizes.
-Shep
ref:
https://github.com/ShepardSiegel/ocpi/blob/master/bin/testBWcommand984
On Fri, Aug 24, 2012 at 1:45 AM, Sutton, Joshua (Contractor)
<[email protected]
<mailto:[email protected]>> wrote:
*UNCLASSIFIED*
Gents,
Are there any OCPI software tools and/or firmware builds to enable
the testing of PCIe throughput?
Also, do you have any OCPI software tools for framework
performance testing?
Cheers,
Josh
----------------------------------------------------
Josh Sutton
Contractor
Signal Processing Consultant
Exactech Pty Ltd
Building 180-34
*Phone:*+ 61 8 7389 5359 <tel:%2B%2061%208%207389%205359>
*M:* 0410 595 477
*Email:*[email protected]
<mailto:[email protected]> , [email protected]
<mailto:[email protected]>
*IMPORTANT*: This email remains the property of the Department of
Defence and is subject to the jurisdiction of section 70 of the
Crimes Act 1914. If you have received this email in error, you are
requested to contact the sender and delete the email.
_______________________________________________
opencpi_dev mailing list
[email protected] <mailto:[email protected]>
http://lists.opencpi.org/listinfo.cgi/opencpi_dev-opencpi.org
_______________________________________________
opencpi_dev mailing list
[email protected]
http://lists.opencpi.org/listinfo.cgi/opencpi_dev-opencpi.org
_______________________________________________
opencpi_dev mailing list
[email protected]
http://lists.opencpi.org/listinfo.cgi/opencpi_dev-opencpi.org
