The hugepages mapping is done in DPDK as part of the library initialization, so this looks like a DPDK issue.
Would you mind reporting that to the dpdk-dev list? On 10/03/2016 11:44, "discuss on behalf of John Wei" <discuss-boun...@openvswitch.org on behalf of johnt...@gmail.com> wrote: >I was trying to run multiple copies of ovs-vswitchd on different >containers on the same host. But, I was not able start them in parallel >because they all seem to try to grab all the available memory in the >beginning, even I have specified --socket-mem > parameter. Is there a work around for this? > > >If I started 2nd ovs-vswitchd after first one is up, release unneeded >memory, then I was able to bring up both. > > > > > >Information on my environment: > >AnonHugePages: 161792 kB >HugePages_Total: 8192 <-- that is 16 GB >HugePages_Free: 7936 <--- each vswitchd used 128 pages >HugePages_Rsvd: 0 >HugePages_Surp: 0 >Hugepagesize: 2048 kB > > >On my 2 core machines, I started vswitched with limit of 256MB > >--socket-mem 128,128 > > >It appears that vswitchd look at that HugePages_Free parameter, grab all >that free memory, then reduce to the limit specified in --socket-mem >option > > >John > > > > > _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
