Overlay network overhead?

Ziti Overlay
1

I am very much still in RTFM, youtube mode.
Lots of great contents and look to me an amazing community forming
.
For low hanging fruit application; migrate traditional CRUD application to be ziti-fied make great sense.

Are there study focus on the overhead of overlay network? in the following category

Application that is sensitive to latency, more so jitter (variance in latency). Consistent long latency but no packet drop; we are ok. High jitter is more problematic. example SIP/Voice Bear traffic (RTP stream).

Application with lots and lots of connections. For example if saw we have millions of MQTTs connection from IOT type of application. Would Ziti make sense here? or something more tailor?

Application required high bandwidth: (streaming video) - 10g, 40g, 100g link. Have anyone done iperf3/ixia type of test to see what overhead on top of underlay network?

10g underlay ethernet link what's the expected real world throughput in ziti overlay?

Are there any hardware acceleration hw test by ziti dev? I assume ziti will be able to take adventage of dpdk/m4, NIC HW/SSL acceleration on the OS stack?

Big part of the design is the underlay network. Would be great to know where the breaking point is. Or we can scale horizontally by throwing HW at it?

Cost effectiveness will be part of the overall equation.
OpenZiti seems to be focus on layer3 and above? no layer 2/lan extension type of construct?

2

Hi @snowman, welcome to the community and to OpenZiti!

I don't know of any studies. We have some data that I usually ask people to do their own, just because in my experience, everyone's network needs are very different. I'll ask around and see if anyone else will offer commentary on the matter.

Not that I know of. We receive bytes, and send bytes so if the network is faster, and if the CPU is fast, I would expect we'll be fine.

Yes, this is the way to scale. You would stand up more than one router as needed. The controller and smart routing will send data down the fastest paths the fabric can calculate. The more routers, the more options for the controller to choose (within reason) are going to help.

Ideally, you would address 'identities' regardless of the underlay network itself, abstracing that complexity away! At that point, there would be no layer 2, 3, 4 etc :wink: That's not reality yet, and we understand that, so we have "tunneling software" that specifically bridges layer 4 to OpenZiti, specifically UDP/TCP. Our tunnelers do nothing with layer 2 at all.