SE310

The following was measured on an SU310-20 (430 RPF, with BS2000 OSD/BC V11.0B and openNet Server V4.0):

Transaction-oriented load

• Maximum transaction rate more than 25,000 TA/s with 100% CPU utilization

• Best mean response time approx. 0.25 milliseconds

Throughput-oriented load for parallel connections

With throughput-oriented mode the CPU utilization depends on the message length. The longer the message, the lower the BS2000 CPU requirement. This dependency does not exist when receiving. Both when sending and receiving, the CPU requirement increases with the number of parallel connections.

The table below shows the maximum throughput and the CPU utilization for 64 KB messages on an SU310-20 for a 1GbE and for a 10 GbE, each with 16 parallel connections and both with standard frames and with jumbo frames.

MTU[bytes]	10 GbE 1500		10 GbE 9000		1 GbE 1500		1 GbE 9000
	Send	Receive	Send	Receive	Send	Receive	Send	Receive
MB/s	1,000	990	850	1,025	112	112	117	118
CPU% for 64KB	100	100	100	100	44	95	27	70

For both the 1 GbE and the 10 GbE a very high utilization can be achieved. For a 10 GbE the BS2000 CPU limits the performance.

Throughput with one connection

• The 1 GbE is fully utilized with one connection both when sending and when receiving: 110-120 MB/s is achieved. The CPU utilization is up to 50% when sending (depending on the message length) and 70% when receiving.

• The following applies for a 10 GbE with one connection:

  • When sending, the throughput depends on the message length. Approximately the following throughput was achieved with both standard frames and with jumbo frames:

    • 250 MB/s with a message length of 8 KB (CPU utilization 77%)

    • 695 MB/s with a message length of 32 KB (CPU utilization 85%)

    • 925 MB/s with a message length of 64 KB (CPU utilization 89%)

  • When receiving, the throughput depends on the message length and amounts to more than 730 MB/s with a CPU utilization of 83% both with standard frames and with jumbo frames.