Ben Pfaff, Hi:
i am clear now. thanks.
i have another two questions to ask.
1. how to limit port's ingress bandwidth?
2. i have moved a port to namespace vm1, and ovs-vsctl set interface
9129f3da-b074-4707-a499-f728e0c8e5e1 ingress_policing_rate=8. but its bandwidth
is not limited by this rule. why?
thanks very much.
At 2017-02-09 00:49:34, "Ben Pfaff" <[email protected]> wrote:
>On Wed, Feb 08, 2017 at 06:27:59PM +0800, lg.yue wrote:
>> Ben Pfaff, Hi:
>> i have to make it clear that QoS here refers to bandwidth
>> limitation(TC), rather than dscp or tos. what i want to do is to limit
>> the bandwidth of vif and ovn's router port.
>> allocate_chassis_queueid and dpif_queue_to_priority tell that queue_id
>> is a incremental number . in kernel function execute_masked_set_action just
>> assign the value of queue_id to skb->priority , then sent it to one queue.
>
>allocate_chassis_queueid() is in ovn-northd, so it's dealing with
>OpenFlow queue IDs (or rather OVN logical queue IDs, but so far that's
>the identity map).
>
>dpif_queue_to_priority() calls into a datapath-specific mapping
>function. For example, the dpif-netlink mapping function (used with the
>Linux kernel datapath) is:
>
> static int
> dpif_netlink_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
> uint32_t queue_id, uint32_t *priority)
> {
> if (queue_id < 0xf000) {
> *priority = TC_H_MAKE(1 << 16, queue_id + 1);
> return 0;
> } else {
> return EINVAL;
> }
> }
>
>execute_masked_set_action() is in the kernel, so anything given to it
>has already been mapped to a sbk_priority.
>
>> my question is:
>> 1. does the flow 'set(skb_priority(0x10002))' have something to do with
>> class htb 1:2 ? and how?
>
>Yes. See the code above.
_______________________________________________
discuss mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
