Op 13/01/2025 om 17:06 schreef Chi vediamo:
Hello Wido, Thank you so much for taking the time to respond.:
I bonded the interfaces for capacity.
I would just create additional BGP sessions and use ECMP balancing for
more capacity.
I was following your YouTube videos and users mail.
Yes, VXLAN seems to be working now. The most dificult/radical part was the
tunning of the parameres on Ubuntu 22. or should I use a different Linux system?
Let me put on a logic it may clear for me; then, with VXLAN:
Wido: Regarding your Youtube video: To me seems there is a VLAN on the TOR. But
makes no sense based on what you mention here. What is the VLAN624 being used
on your Youtube video @time 24:10 https://www.youtube.com/watch?v=X02bxtIC0u4
VLAN is a Cumulus naming. These are all VXLAN VNI. In this case I showed
the example how Cumulus would be the gateway for VNI 624
Then, Management servers will be on the same vxlan as kvm cloudbr1 then this
should be the management vxlan. Correct?
Storage I did put the ceph on vxlan2 with BGP; then, should I removed from
vxlan ior keep it on the management vxlan? which one should be the best
practice.?
Storage doesn't need to be in a VNI/VXLAN. It just needs to be reachable
by the hypervisor. Plain L3 routing, nothing special. Just make sure
both can ping eachother.
This is really just basic routing.
What About Network planning with Cloudstack for NFVs and low latency:
Final Question. Does Cloudstack supports Sriov ? Any documentation besides an
old document that inidicates we should list the PCI interfaces ? IS there ANy
documentation about this?
No idea :-)
I was thinnking SRIOV will align better with VXLAN/EVPN
Or do you recommend to use DPDk ?
If DPDK wihich one will you recommend,
- OpenVswitch - if this is selected then I will have to create the Bridges
with Openvswitch, and should work with VXLAN/EVPN , right?
- Or tunsgteen(OpenSDN) - I know Contrail pretty much very well, Not sure if
there is any documentation about OpenSDN - Formerly Tungteen - with Cloudstack
and VXLAN/EVPN,
- Seems VXLAN/EVPN and tungsteen are mutually exclusive, plus not supported on
Ubuntu 22 as of yet, are both of this statements correct?
Thank you so much for taking the time to respond.
Tata Y.
On Jan 13, 2025, at 8:45 AM, Wido den Hollander <w...@widodh.nl> wrote:
Op 11/01/2025 om 00:36 schreef Chi vediamo:
Forgot to add Wido
The Isolation should start, and should happen at the HOST running KVM, No VLANS
needed at the Host, neither at the Leafs.
But seems isolation is not happening! using cloudstack 4.20, maybe is a detail
missing on my side.
What Cloudstac documentation mean with "On the Spine router(s) the VNIs will
terminate and they will act as IPv4/IPv6 gateways"
DO "Need" vlans to be configured on the Leafs. I will understand if the VM
isolation is based on VLANS that will be needed, but if the VM Isolation is going to be
VXLAN
what for I will need the VLANS at the leafs?
No VLANs needed. Forget Guest and Storage traffic. The hypervisor will do
everything routed! It will talk BGP with the Top-of-Rack.
You will reach your storage via routing without the need of VXLAN, have it
route through your network.
No need for bonding on your hypervisor either, it just has two BGP uplink
sessions towards the ToR.
Just connect your storage to your network and make it reachable via L3, same
goes for the management server.
But before starting to use CloudStack with VXLAN, make sure you get VXLAN
working without CloudStack so you really understand how VXLAN works.
Wido
based on this scenario from Guido presso:
[spine1] [spine2] no ibgp between spines
/ /
/ / evpn/bgp unnumbered
/ /
/ /
[leaf1]-----ibgp---[Leaf2]
\ /
\ /
\ /
\ /
| bond1 bond2 |
| [cloudbr1] |
| vxlan1 |
| vxlan2 |
| Host 1 - KVM |
| SRIOV |
--------------------
the vxlan1 will handle the guest traffic
the vxlan2 will handle main storage
I have management over a regular vlan for now but will create a vxlan3 for it
and vxlan2000 for Public traffic.
vxlan1, vxlan2, vxlan3, and vxlan2000 are goint to be created manually or
Cloudstack should/will create them when I choose the VXLAN for each one of the
traffic names?
Or should I use VXLAN for the Guest traffic only and use regular vlan isolation
for the the remaining - management, storage and public ?
I have 4 physical hosts, and i am using same ip addressing on vxlan 1 and vxlan
2 connect to different leafs and i am still able to ping between them.
Thank you
Tata Y.
On Jan 6, 2025, at 9:08 PM, Chi vediamo <tatay...@gmail.com> wrote:
Hello Community,
First: Thank you Wido, your answers on previous emails to the community help a
lot. I read the vincent.bernat document, but his example uses VLAN mapping at
the Switch level.
I was thinking to use the LEAF-SPINE as a transport only , and the seetings on
the Host will take care of the Isolation.
But is not working that way. Should I create the traditional VXLAN,
VLAN/VNI/VRF on the LEAF Switches to properly isolate it?
We are using SONIC NOS community version, nothing fancy.
The BGP unnumbered evpn etc works fine.
The output:
vtysh -c 'show interface vxlan2'
VNI: 2
Type: L2
Tenant VRF: default
VxLAN interface: vxlan2
VxLAN ifIndex: 14
SVI interface: storage0
SVI ifIndex: 13
Local VTEP IP: 172.2.0.60
Mcast group: 0.0.0.0
Remote VTEPs for this VNI:
172.2.0.59 flood: HER
172.2.0.32 flood: HER
172.2.0.30 flood: HER
172.2.0.28 flood: HER
172.2.0.26 flood: HER
bridge fdb show dev vxlan2
8a:be:71:4c:e0:20 vlan 1 extern_learn master storage0
8a:be:71:4c:e0:20 extern_learn master storage0
b2:33:bb:84:cc:38 vlan 1 extern_learn master storage0
b2:33:bb:84:cc:38 extern_learn master storage0
86:07:90:2b:db:db vlan 1 extern_learn master storage0
86:07:90:2b:db:db extern_learn master storage0
4a:28:60:90:76:42 vlan 1 extern_learn master storage0
4a:28:60:90:76:42 extern_learn master storage0
22:d6:49:9f:08:07 vlan 1 extern_learn master storage0
22:d6:49:9f:08:07 extern_learn master storage0
fe:4a:fb:63:9d:3a vlan 1 extern_learn master storage0
fe:4a:fb:63:9d:3a extern_learn master storage0
ee:78:b4:d8:3f:a0 vlan 1 master storage0 permanent
ee:78:b4:d8:3f:a0 master storage0 permanent
00:00:00:00:00:00 dst 172.2.0.24 self permanent
00:00:00:00:00:00 dst 172.2.0.26 self permanent
00:00:00:00:00:00 dst 172.2.0.28 self permanent
00:00:00:00:00:00 dst 172.2.0.30 self permanent
00:00:00:00:00:00 dst 172.2.0.32 self permanent
00:00:00:00:00:00 dst 172.2.0.59 self permanent
fe:4a:fb:63:9d:3a dst 172.2.0.24 self extern_learn
vtysh -c 'show interface vxlan1'
Interface vxlan1 is up, line protocol is up
Link ups: 1 last: 2025/01/06 23:53:01.17
Link downs: 1 last: 2025/01/06 23:53:01.17
vrf: default
index 14 metric 0 mtu 9050 speed 4294967295
flags: <UP,BROADCAST,RUNNING,MULTICAST>
Type: Ethernet
HWaddr: ea:d3:68:02:7d:f7
inet6 fe80::e8d3:68ff:fe02:7df7/64
Interface Type Vxlan
Interface Slave Type None
VxLAN Id 100 VTEP IP: 10.23.13.14 Access VLAN Id 1
protodown: off
vtysh -c 'show evpn vni 1'
VNI: 1
Type: L2
Tenant VRF: default
VxLAN interface: vxlan1
VxLAN ifIndex: 14
SVI interface: cloudbr1
SVI ifIndex: 12
Local VTEP IP: 10.23.13.14
Mcast group: 0.0.0.0
No remote VTEPs known for this VNI
Number of MACs (local and remote) known for this VNI: 0
Number of ARPs (IPv4 and IPv6, local and remote) known for this VNI: 0
Advertise-gw-macip: No
Advertise-svi-macip: No
and I can ping the IPV6 that is routed using the FRR from :60 which is in VXLAN
2 to :14 which is in VXLAN 1
ping -I 20XX:5XX:56XX:fff0::2:60 20XX:5XX:56XX:fff0:0:2:13:14
PING 20XX:5XX:56XX:fff0:0:2:13:14(20XX:5XX:56XX:fff0:0:2:13:14) from
20XX:5XX:56XX:fff0::2:60 : 56 data bytes
64 bytes from 20XX:5XX:56XX:fff0:0:2:13:14: icmp_seq=1 ttl=61 time=0.293 ms
64 bytes from 20XX:5XX:56XX:fff0:0:2:13:14: icmp_seq=2 ttl=61 time=0.222 ms
Then, my questions are:
are you using at the Leaf Switches/routers a regular Mapping VLAN to VNI VXLAN
with VRF ? IF not, Can you share a FRR config of your Switches?
Or should I use an enterprise SONIC switch software ?
What other possibilities are with the modifyvxlan.sh That Wido states on some
user mails.
Thank you
Tata Y.
