Description: Today there is a subset of /24 subnet allocated for all VIRL simulations, split equally across 3 servers, 74 /32 addresses per server as follows:
- tb4-virl1 - {"start": "10.30.51.31", "end":"10.30.51.105"}
- tb4-virl2 - {"start": "10.30.51.106", "end":"10.30.51.180"}
- tb4-virl3 - {"start": "10.30.51.181", "end":"10.30.51.254"}
Each CSIT simulation takes 5 addresses (mgmt1, mgmt2, tg, sut1, sut2), each csit-vpp and vpp-csit verify job uses 3 simulation to parallize test execution and reduce execution time, hence each verify virl job takes 3*5=15 ipv4 mgmt addresses. This means today each server has capacity to run concurrently up to only 4 verify jobs concurrently (4*3*5=60), for total of 3*4=12 concurrent jobs across 3 virl servers. Once Centos7 tests productized, where two jobs are always executed in parallel, this will reduce it down to 2 concurrent jobs per server. Not good. It's basically a show stopper to productize Centos7 into vpp-csit-verify per patch jobs.
Solution: Need to increase IPv4 address space given to VIRL hosts. Getting a dedicated /24 subnet per VIRL server with 250 usable addresses (instead of 74 available today), would give address capacity for 50 concurrent simulations. With three servers that's 3*50=150 simulations, increasing concurrency limit to 150/3=50 jobs. This should work for CSIT and VPP projects, and enable Centos7 per patch tests.