Wired Intelligent Edge

One of the key features of Aruba 8320 is the performance "2.5Tbps with 1.905MPPS", what Tbps used for and what MPPS used for?

Tbps - Tera Bits Per Second which means Million megabit per second, this rating measures the maximum bandwidth can pass through the switch across all ports. While MPPS - Mega Packet  Per Second where mega stand for Million, this rating measures how many million packets (minimum packet size) the switch able to process per second.

In general, every 1 Gbps Ethernet requires 1.488 Mpps and the math behind it: 

1Gbps = 1,000,000,000bps/8bit = 125,000,000

125,000,000 / (64+8 +12) = 1,488Mpps

64 bytes per packet as the standard average packet size and 8 bytes for the header. The 12 bytes are for the frame interval.

Please note the PPS value is variable and depend on the packet size here we assumed the packet size is 84 bytes in total.

Sorry, but what is your question?

As you mentioned, T is associated with switching and M is associated with packets per second.

Tbps is a measure of the performance of the backplane while MPPS is a measure of the performance of the routing or switching of the equipment.  In your example you assume that packets are sent with the smallest size possible for Ethernet in which case the MPPS limit would govern the total throughput the switch can process.  The specs for the 8320 are actually 2.5Tbps 1.905 TPPS or 1905 MPPS so it could process over 1.28016 Tbps of the 84 byte minimum possible size frames.  Considering the fact that a standard MTU is 1500 bytes then there is a good chance that the backplane of the switch will become the bottleneck before the limit of the switch to process the packets in the real world where the average TCP packet is 300-500 bytes.

It may be worth noting that this is a more technical way of saying non-blocking since if all 48x 10G and 6x 40G ports were sending traffic then that totals to .720 Tbps.  Even with every port running at 100% utilization with the smallest possible frame size allowed for Ethernet the switch is not the bottleneck.

Just an additional note. Even  new 8320 (32x 40G) is line rate:

Consider 32x40G inbound traffic going to 32x40G outbound means

(1,280Gbps / 8) / (64+8+12) = 1,905 Mpacket/s.