[dpdk-users] DPDK Linux kernel version requirement

James Huang jamsphon at gmail.com
Thu Nov 17 20:46:02 CET 2016


I did a bunch test with DPDK v16.07

- system (netgate appliance box)
CentOS release 6.5 (Final)
Linux NETGATE-TEST-Appliance 2.6.32-431.5.1.el6.x86_64 #1 SMP Wed Feb
12 09:03:27 CET 2014 x86_64 x86_64 x86_64 GNU/Linux
4 CPU cores @2.4GHz
8 GB memory
6 1Gb NICs, eth0 - eth5

- install module (x86_64)
igb_uio.ko
rte_kni.ko

- reserve huge pages memory (2M*512=1GB)
mkdir -p /mnt/huge
mount -t hugetlbfs nodev /mnt/huge
echo 512 > /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages

- run test programs
testpmd.i686 -c7 -n3 --vdev=eth_pcap0,iface=eth4
--vdev=eth_pcap1,iface=eth5 -- -i --nb-cores=2 --nb-ports=2
--total-num-mbufs=2048
packet size 64B
Rx-pps and Tx-pps around 120K, or 77Mbps

testpmd.i686 -c7 -n3 -- -i --nb-cores=2 --nb-ports=2 --total-num-mbufs=2048
use igb_uio driver
packet size 64B
Rx-pps and Tx-pps around 1.4M, or 900Mbps

./pktgen.i686 -c f -n 4 -- -p 0x3 -P -m "1.0, 2.1" -N -T
packet size from 64 to 1500, all throughputs could reach 998-999Mbit/s


According to the test results, could I say DPDK 32bit applications are
GOOD to run on Linux 64bit kernel v2.6.32?

Regards,
James Huang


On Thu, Nov 17, 2016 at 10:32 AM, Wiles, Keith <keith.wiles at intel.com> wrote:
>
>> On Nov 17, 2016, at 12:27 PM, James Huang <jamsphon at gmail.com> wrote:
>>
>> Hi,
>>
>> Nice to join DPDK community.
>>
>> read from DPDK system requirements for Linux
>> (http://dpdk.org/doc/guides/linux_gsg/sys_reqs.html)
>>
>>  Kernel version >= 2.6.34
>>
>> I'd like to know if there is any reason to block DPDK build and run on
>> older kernel version, i.e. 2.6.32, as carried on CentOS 6. where
>> gcc=4.4.
>>
>> Could we just keep our CentOS 6 64bit x86_64 OS and continuing running
>> 32bit applications with DPDK injected.
>>
>> or, there are good reasons *must* upgrade to CentOS 7, where
>> kernel=3.10.0, gcc=4.8, and re-build applications to 64bit as well?
>
> The only issues I can think of are no huge page support and the igb_uio or kernel components may not build. Having no huge page support can be a performance hit, but you should use —no-huge option. I know some people have it building on older kernels, but it has been many revs back.
>
>>
>> Thanks a lot!
>>
>>
>> James
>
> Regards,
> Keith
>


More information about the users mailing list