Where is the performance bottleneck in Connection Server?

[mren]

Lots of edge devices update a temperature property to ThingWorx platform via Connection Sever(s) every 5 secs. When I use one Connection Server, ThingWorx platform can process 3k value steams per sec, and the increment of the value streams processing per sec is about 2-3k:

#	Condition	Value Stream processing per sec	Connection Server			ThingWorx Platfrom
			max CPU (%)	network i/o recv(MB/s)	disk I/O write (MB/s)	max CPU (%)	network i/o recv(MB/s)	disk I/O write (MB/s)
1	2 CxS + 1 TWX	5k	48	2.5	1.5	15	4.5	130
2	3 CxS + TWX	8k	60	2.5	1	18	5.5	175
3	4CxS + TWX	10k	60	3	1	18	6	190

As you've seen above, the performance utilization of Connection Server hasn't been reached to the bottleneck. Where should be the performance bottleneck in Connection Server?

Thanks,

Ren

[ttielebein]

What exactly are you saying is the problem?

[mren]

I'm sorry that I didn't describe it exactly.

To connect more edge devices and accept their properties updating requests, we have to use more connection server. My question is, why I have to add another connection server to perform more things' connections/value streams, but the connection server we are using hasn't reach the bottom of the hardware(CPU, memory, etc) utilization? As the table described, the connection server's CPU max utilization is about 50%-60%(transient peak), average utilization is about 30%, and the bandwidth of network/disk can be ignored.

Thanks for your reply.