Hi Robert,

Yes that is correct about RSC vs. LRO. Have you tried turning of RSC on 82599? 
ethtool -K lro off ethX

Also how about if you disable DCB? Does that have impact on latency?

Also could you please send us following after loading the driver:

Output from:

dmesg
lspci -vvv


Thanks,
Stephen

-----Original Message-----
From: Torres, Robert (CCL) [rtor...@carnival.com<mailto:rtor...@carnival.com>]
Sent: Thursday, November 08, 2012 03:04 PM Pacific Standard Time
To: 'e1000-de...@lists.sf.net'
Subject: [E1000-devel] Poor Read Performance with 82599EB on Red Hat Linux


Getting two distinctive behaviors from two RHEL 5 servers. Seems the host with 
the 82599 NICs reports significant latency when read requests surpass 32KB. The 
host with the 82598 NICs does not demonstrate this behavior. It's our 
understanding that the two NIC adapters perform different offload capability.  
The 82599-based adapters support HW based receive side coalescing (RSC) which 
can merge multiple frames from the same IPv4 TCP/IP flow into a single 
structure that can span one or more descriptors, while the 82598-based adapter 
rely on driver side LRO for aggregation.



DCB is enabled on the arrays ports and don't know if this has any relevance to 
the above issue.



------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_nov
_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel&#174; Ethernet, visit 
http://communities.intel.com/community/wired

Reply via email to