-
Notifications
You must be signed in to change notification settings - Fork 51
Description
Hi,
I am running SwitchML allreduce_benchmarks on a cluster of nodes with a mix of MLX5 NICs and some with Intel 82599 ES 10G NICs thus I'm using DPDK as the communication backend. I need to share the NIC on each host with other traffic so I'm virtualizing it by creating a virtual function of the PCI device in order to use the original device for general purpose traffic and the virtual device to run the SwitchML app. However, when I try to run SwitchML with the virtual device, I'm getting the following error:
Submitting 5 warmup jobs.
EAL: Detected 20 lcore(s)
EAL: Detected 2 NUMA nodes
EAL: Multi-process socket /var/run/dpdk/rte/mp_socket
EAL: Selected IOVA mode 'VA'
EAL: No available hugepages reported in hugepages-1048576kB
EAL: Probing VFIO support...
EAL: VFIO support initialized
EAL: PCI device 0000:01:10.1 on NUMA socket 0
EAL: probe driver: 8086:10ed net_ixgbe_vf
EAL: using IOMMU type 1 (Type 1)
E1116 19:08:30.807562 74629 dpdk_master_thread_utils.inc:277] Flow isolated mode failed: 1 Function not implemented
F1116 19:08:31.361418 74629 dpdk_master_thread_utils.inc:154] Flow rule can't be added: 1Function not implemented
*** Check failure stack trace: ***
@ 0x7f2291e280cd google::LogMessage::Fail()
@ 0x7f2291e29f33 google::LogMessage::SendToLog()
@ 0x7f2291e27c28 google::LogMessage::Flush()
@ 0x7f2291e2a999 google::LogMessageFatal::~LogMessageFatal()
@ 0x564504203fb2 switchml::InsertFlowRule()
@ 0x5645042048ca switchml::InitPort()
@ 0x564504205ae6 switchml::DpdkMasterThread::operator()()
@ 0x7f2291ae34c0 (unknown)
@ 0x7f22915766db start_thread
@ 0x7f229015761f clone
which seems to be caused by struct rte_flow_error error; LOG_IF(FATAL, rte_flow_validate(port_id, &attr, pattern, action, &error) != 0) << "Flow rule can't be added: " << error.type << (error.message ? error.message : "(no stated reason)"); in InsertFlowRule function. Any ideas on why this is happening and whether I can overcome this? Much appreciate it.
Thank you.