Hi,
We are measuring packet latency on 10G port pair using
bi-directional UDP 64 byte packets.
Test machine is connected with load generator using direct 10G ports.
Test machine configuration:
CPU: 2 - 6 core Intel(R) Xeon(R) CPU E5645 @
2.40GHz (Total 12 cores)
Memory: 6 GB
10G ports: 2 82599 SFP - lspci ID (0x10FB)
kernel: Vanila-2.6.27.45
ixgbe: 3.8.21
ixgbe driver is loaded with 6 Receive queues per NIC (each is
binded with a specific core) and Dynamic ITR. We are measuring routing
performance so no other network modules are loaded.
We are generating load at 10G (bi-direction 20G) using 64byte UDP
packets and getting throughput around 4 Gbps with per packet latency
around 1.5 millisecond. (ksoftirqd is taking 99% cpu on few cores) So,
seems to be pps rate is too high for the machine.
On the other side, when we generate load at 2G (bi-direction 4G)
using 64byte UDP packets, latency is reduced to 50 microseconds and
ksoftirqd is also under control.
Is there any way/configuration through which we can keep the low
latency (around 100-150 microseconds) even at high incoming rate of
packet even with packet drops ?
Rgds,
Nishit Shah
------------------------------------------------------------------------------
Keep yourself connected to Go Parallel:
VERIFY Test and improve your parallel project with help from experts
and peers. http://goparallel.sourceforge.net
_______________________________________________
E1000-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel® Ethernet, visit
http://communities.intel.com/community/wired
