ODL Helium SR3 switch scalability results

In this section we present indicative results regarding the switch scalability of OpenDaylight Helium SR3 controller with Cbench and MT-Cbench switches, both in idle and active tests. In the first case we investigate the maximum size of switch topology that can be connected to the controller. In the second case we investigate how controller performance varies as topology scales.

Idle test results (Cbench vs MT-Cbench)

• Configuration 1:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Helium SR3 ("RPC drop-test" mode)
  • Generator: Cbench ("Latency" mode)
  • Switches per thread: [50,100,200,400,800,1000,1500,2000,2500,3000,3500,4000,4500,5000]
  • Threads: 1
  • Thread creation delay: 0

• Configuration 2:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Helium SR3 ("RPC drop-test" mode)
  • Generator: MT-Cbench ("Latency" mode)
  • Switches per thread: 50
  • Threads: [1,2,4,8,16,20,30,40,50,60,70,80,90,100]
  • Thread creation delay: [500,1000,2000,4000,8000,16000]
  • Generator mode: Latency

• Configuration 3:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Helium SR3 ("RPC drop-test" mode)
  • Generator: MT-Cbench ("Latency" mode)
  • Switches per thread: 50
  • Threads: [1 2,4,8,16,20,30,40,50,60,70,80,90,100,120,140,160,180,200,220,240,260,280,300]
  • Thread creation delay: [500,1000,2000,4000,8000,16000]
  • Generator mode: Latency

• Conclusions:
  • with MT-Cbench we were able to successfully boot 5000 switches at ~400 seconds best time. We might be able to boot switches beyond this number but we haven't tried this yet.
  • with Cbench we were able to successfully boot up to 400 switches only

Active test results (Cbench vs MT-Cbench, Helium vs Hydrogen)

• Configuration 1:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Helium SR3 ("RPC drop-test" mode)
  • Generator: Cbench ("Latency" mode)
  • Switches per thread: [50,100,200,400,800,1000,1500,2000,2500,3000,3500,4000,4500,5000]
  • Threads: 1
  • Thread creation delay: 0
  • External test repeats: 2

• Configuration 2:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Helium SR3 ("RPC drop-test" mode)
  • Generator: MT-Cbench ("Latency" mode)
  • Switches per thread: [50]
  • Threads: [1,2,4,8,16,20,30,40,50,60,70,80,90,100]
  • Thread creation delay: 15000
  • External test repeats: 2

• Configuration 3:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Helium SR3 ("RPC drop-test" mode)
  • Generator: MT-Cbench ("Latency" mode)
  • Switches per thread: [50]
  • Threads: [1,2,4,8,16,20,30,40,50,60,70,80,90,100,120,140,160,180,200,220,240,260,280,300]
  • Thread creation delay: 15000
  • External test repeats: 2

• Configuration 4:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Hydrogen ("RPC drop-test" mode)
  • Generator: Cbench ("Latency" mode)
  • Switches per thread: [50,100,200,400,800,1000,1500,2000,2500,3000,3500,4000,4500,5000]
  • Threads: 1
  • Thread creation delay: 0
  • External test repeats: 2

• Configuration 5:
  • 32-core server, 256GB RAM
  • Controller: OpenDaylight Hydrogen ("RPC drop-test" mode)
  • Generator: MT-Cbench ("Latency" mode)
  • Switches per thread: [50]
  • Threads: [1,2,4,8,16,20,30,40,50,60,70,80,90,100]
  • Thread creation delay: 15000
  • External test repeats: 2

• Conclusions
  • Again, with MT-Cbench we were able to successfully get performance results for 5000 switches (without having tried for more), while with Cbench we got non-zero results for up to 400 switches.
  • In Hydrogen, MT-Cbench managed to successful generate traffic with up to 5000 switches. However, performance was almost 0.5x compared to Helium.