[largescale-sig][neutron] What driver are you using?
Hey large-scalers, We had a discusion in my company (OVH) about neutron drivers. We are using a custom driver based on BGP for public networking, and another custom driver for private networking (based on vlan). Benefits from this are obvious: - we maintain the code - we do what we want, not more, not less - it fits perfectly to the network layer our company is using - we have full control of the networking stack But it also have some downsides: - we have to maintain the code... (rebasing, etc.) - we introduce bugs that are not upstream (more code, more bugs) - a change in code is taking longer, we have few people working on this (compared to a community based) - this is not upstream (so not opensource) - we are not sharing (bad) So, we were wondering which drivers are used upstream in large scale environment (not sure a vlan driver can be used with more than 500 hypervisors / I dont know about vxlan or any other solution). Is there anyone willing to share this info? Thanks in advance!
I feel like it depends a lot on the scale/target usage (public vs private cloud). But at $dayjob, we are leveraging - vlans for external networking (linux-bridge + OVS) - vxlans for internal Openstack networks. We like the simplicity of vxlan with minimal overlay configuration. There are some scaling/performance issues with stuff like l2 population. VLANs are okay but it's hard to predict the next 5 years of growth. On Mon, May 10, 2021 at 8:34 AM Arnaud Morin <arnaud.morin@gmail.com> wrote:
Hey large-scalers,
We had a discusion in my company (OVH) about neutron drivers. We are using a custom driver based on BGP for public networking, and another custom driver for private networking (based on vlan).
Benefits from this are obvious: - we maintain the code - we do what we want, not more, not less - it fits perfectly to the network layer our company is using - we have full control of the networking stack
But it also have some downsides: - we have to maintain the code... (rebasing, etc.) - we introduce bugs that are not upstream (more code, more bugs) - a change in code is taking longer, we have few people working on this (compared to a community based) - this is not upstream (so not opensource) - we are not sharing (bad)
So, we were wondering which drivers are used upstream in large scale environment (not sure a vlan driver can be used with more than 500 hypervisors / I dont know about vxlan or any other solution).
Is there anyone willing to share this info?
Thanks in advance!
Hi Laurent, Thanks for your reply! I agree that it depends on the scale usage. About the VLAN you are using for external networks, do you have/want to share the number of public IP you have in this L2 for a region? Cheers, On 11.05.21 - 19:21, Laurent Dumont wrote:
I feel like it depends a lot on the scale/target usage (public vs private cloud).
But at $dayjob, we are leveraging
- vlans for external networking (linux-bridge + OVS) - vxlans for internal Openstack networks.
We like the simplicity of vxlan with minimal overlay configuration. There are some scaling/performance issues with stuff like l2 population.
VLANs are okay but it's hard to predict the next 5 years of growth.
On Mon, May 10, 2021 at 8:34 AM Arnaud Morin <arnaud.morin@gmail.com> wrote:
Hey large-scalers,
We had a discusion in my company (OVH) about neutron drivers. We are using a custom driver based on BGP for public networking, and another custom driver for private networking (based on vlan).
Benefits from this are obvious: - we maintain the code - we do what we want, not more, not less - it fits perfectly to the network layer our company is using - we have full control of the networking stack
But it also have some downsides: - we have to maintain the code... (rebasing, etc.) - we introduce bugs that are not upstream (more code, more bugs) - a change in code is taking longer, we have few people working on this (compared to a community based) - this is not upstream (so not opensource) - we are not sharing (bad)
So, we were wondering which drivers are used upstream in large scale environment (not sure a vlan driver can be used with more than 500 hypervisors / I dont know about vxlan or any other solution).
Is there anyone willing to share this info?
Thanks in advance!
We are "lucky" that external connectivity needs are limited. We have between 50-100 IP per L2 usually. We do not have huge pools of public IPs which are harder to handle/scale as with a public cloud. On Mon, May 17, 2021 at 10:59 AM Arnaud Morin <arnaud.morin@gmail.com> wrote:
Hi Laurent,
Thanks for your reply! I agree that it depends on the scale usage. About the VLAN you are using for external networks, do you have/want to share the number of public IP you have in this L2 for a region?
Cheers,
On 11.05.21 - 19:21, Laurent Dumont wrote:
I feel like it depends a lot on the scale/target usage (public vs private cloud).
But at $dayjob, we are leveraging
- vlans for external networking (linux-bridge + OVS) - vxlans for internal Openstack networks.
We like the simplicity of vxlan with minimal overlay configuration. There are some scaling/performance issues with stuff like l2 population.
VLANs are okay but it's hard to predict the next 5 years of growth.
On Mon, May 10, 2021 at 8:34 AM Arnaud Morin <arnaud.morin@gmail.com> wrote:
Hey large-scalers,
We had a discusion in my company (OVH) about neutron drivers. We are using a custom driver based on BGP for public networking, and another custom driver for private networking (based on vlan).
Benefits from this are obvious: - we maintain the code - we do what we want, not more, not less - it fits perfectly to the network layer our company is using - we have full control of the networking stack
But it also have some downsides: - we have to maintain the code... (rebasing, etc.) - we introduce bugs that are not upstream (more code, more bugs) - a change in code is taking longer, we have few people working on this (compared to a community based) - this is not upstream (so not opensource) - we are not sharing (bad)
So, we were wondering which drivers are used upstream in large scale environment (not sure a vlan driver can be used with more than 500 hypervisors / I dont know about vxlan or any other solution).
Is there anyone willing to share this info?
Thanks in advance!
participants (2)
-
Arnaud Morin
-
Laurent Dumont