router-perf: update default CLIENTS after maxconn changed from 20k to 50k in 4.12

[qiliRedHat]

Connection Capacity for default 2 routers
number of router(default 2) x default maxConnections(50k) = 100k connections
Culculation for proper CLIENT
small scale(100 routes)
For http and passthrough, number of routes(100 for large scale) x clients(1000) = 100k.
For edge and re-encrypt, number of routes(100 for large scale) x clients(500)x2 = 100k
To leave some buffer, 400 is a good number for SMALL_SCALE_CLIENTS

For mix termination, number of routes(100) x 2 (http and passthrough) x clients(150) + routes(100) x 2(edge and re-encrypt) x 2(connection per termination) x clients(150) =90k connections
150 is a good number for SMALL_SCALE_CLIENTS_MIX

large scale(500 routes)
For http and passthrough, number of routes(500 for large scale) x clients(200) = 100k
For edge and re-encrypt, number of routes(500 for large scale) x clients(100)x2 = 100k
To leave some buffer, 80 is a good number for LARGE_SCALE_CLIENTS

For mix termination, number of routes(500) x 2 (http and passthrough) x clients(30) + routes(500) x 2(edge and re-encrypt) x 2(connection per termination) x clients(30) =90k connections
30 is a good number for LARGE_SCALE_CLIENTS_MIX

Proposal:

SMALL_SCALE_CLIENTS 1 40 200-> 1 200 400
SMALL_SCALE_CLIENTS_MIX 1 20 80 -> 1 50 150
LARGE_SCALE_CLIENTS 1 20 80-> 1 40 80
LARGE_SCALE_CLIENTS_MIX 1 10 20-> 1 10 30

@rsevilla87 Please let me know your comments. I will open a PR when we get an agreement.

[rsevilla87]

I like this proposal @qiliRedHat, however it comes with some caveats since maxConnections=50K is defaulted in 4.12 as you mentioned. Additional info at https://issues.redhat.com/browse/NE-1072

If I've understood correctly, the maximum connections created by each test should be :

Edge & passthrough = routes * clients * 2 (edge & passthrough)
Mix = routes * clients * 1.5 (half of the routes are edge & passthrough)

• Small scale -> 100x400x2=80K

• Small scale mix -> 400x150x1.5=90K <--- Leaving only 10K connections as buffer could be dangerous, I propose using 125 clients to reach 75K connections

• Large scale -> 500x80x2=80K

• Large scale mix -> 2000x30x1.5=90K <--- Leaving only 10K connections as buffer could be dangerous, I propose using 25 clients to reach 75K connections

Where each router supports up to 50K connections, we deploy 2 routers by default. Hence they can handle up to 100K connections.

Other than that, I'd remove some of the iterations to make this test a bit shorter as proposed already at #452

SMALL_SCALE_CLIENTS 1 40 200-> 1 400
SMALL_SCALE_CLIENTS_MIX 1 20 80 -> 1 125
LARGE_SCALE_CLIENTS 1 20 80-> 1 80
LARGE_SCALE_CLIENTS_MIX 1 10 20-> 1 25

[qiliRedHat]

I made the change in PR https://github.com/cloud-bulldozer/e2e-benchmarking/pull/518/files

[rsevilla87]

Fixed in #519