Hello,

I have a case where a very slight difference in packet sizes is causing the
driver to switch from the "low_latency" RX interrupt rate to the
"bulk_latency" one.

I'm trying to check if this behavior is expected. I appreciate any
help/explanation. Please let me know if this is not the correct mailing
list for this question.

>From an external traffic generator, I send to the Linux system udp packets
of varying sizes (with dummy udp payload) but with one (the same) exact
UDP/IP header reused over and over. Below are some chosen cases  of traffic
patterns that result in significantly different RX interrupt rate at the
Linux system despite very close values of average packet sizes as well as
frames/sec rate.

High (~20k/sec) interrupt rate cases:
1. Frame sizes used are 571 bytes and 572 bytes. The traffic load is
a uniform mixture of the 2 sizes; the average size is 571.5 bytes
2. Frames are all 571 bytes long.
3. Frame sizes used are 571 bytes, 572 bytes and 573 bytes. The traffic
load is a uniform mixture of the 3 sizes; the average size is 572 bytes.

Low (~4k/sec) interrupt rate cases:
1. Frame sizes used are 570 bytes, 571 bytes and 572 bytes. The traffic
load is a uniform mixture of the 3 sizes. The average size is 571.
2. Frame sizes used are between 570 and 574.  The traffic load is a uniform
mixture of the 5 sizes. The average size is 572.

The interrupt rate was calculated/averaged over 100 seconds for each of the
cases (based on /proc/interrupts output). The frames/sec rate in all of the
cases was around 209k/sec.

The driver and kernel versions are below:
# ethtool -i eth4
driver: igb
version: 3.0.6-k2
firmware-version: 3.19-0
bus-info: 0000:07:00.3

#uname -a
Linux abcdefg 2.6.32-131.0.15.el6.x86_64 #1 SMP Tue May 10 15:42:40 EDT
2011 x86_64 x86_64 x86_64 GNU/Linux

Thanks,
Al.
ps. in case it matters, the packets are routed through the Linux system
(not destined to it). They enter through eth4 (in the above example) and
exist through a different eth port. eth4 is only used to receive packets.
------------------------------------------------------------------------------
Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex
infrastructure or vast IT resources to deliver seamless, secure access to
virtual desktops. With this all-in-one solution, easily deploy virtual 
desktops for less than the cost of PCs and save 60% on VDI infrastructure 
costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox
_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel® Ethernet, visit 
http://communities.intel.com/community/wired

Reply via email to