more packet loss when transmitting 64byte frame size than 1518byte

[soginy]

i am carrying some test on network switches using spirent traffic generator i am basically using two host to generate traffic in order to congest the network. both host generate equal amount of traffic. for instance i set both host to generate 50000Mb on traffic so total traffic will be 100. i increase the amount of traffic generated gradually to see when packets starts getting drop. i observed that when ran the test with 64byte frame packets starts getting drop quicker than when i carried out the same test using 1518 byte frame. can anyone explain why it is so ? i have carried out this test on two different types of switches and it was same result. here is the link to the result. https://www.dropbox.com/s/igsnmj4g87vxnzm/results.xlsx?dl=0 column S is the frame loss column in the spreedsheet.

thanks

[Reza Sharifi]

Are you referring to the out-of-sequence column?

What type of switches are these? 

HTH

[soginy]

Yes the out-of-sequence column. I actually searched online for reports of the same testing. and it was same behaviour i just need the explanation as to why its so. I have use a cisco 3560 switch and an Hp-aruba M3810 switch for the test with same behaviour. i attached one such result. you can search for rfc2544 test. 

[sathvik k v]

Which switch models are you testing? which column are you referring to as packet loss?

Usually larget sized packets are more prone to packet loss than smaller sized packet. Larger the size of the packets the fewer the packets that can be queued in the buffer. Thus large sized packets more easily fill the buffer than small sized packets, and cause packet drop. Large packets also take longer to be received, queued, and transmitted than small packets.

-Sathvik

[soginy]

Which switch models are you testing? which column are you referring to as packet loss?

Usually larget sized packets are more prone to packet loss than smaller sized packet. Larger the size of the packets the fewer the packets that can be queued in the buffer. Thus large sized packets more easily fill the buffer than small sized packets, and cause packet drop. Large packets also take longer to be received, queued, and transmitted than small packets.

Yes the out-of-sequence column. I actually searched online for reports of the same testing. and it was same behaviour i just need the explanation as to why its so. I have use a cisco 3560 switch and an Hp-aruba M3810 switch for the test with same behaviour. i attached one such result. you can search for rfc2544 test.  I also thought larger size packets should be more prone to drops but it seems for the smaller packets the switch buffers dont even get filled before the packets start to drop. I am thinking there is a maximum number of frames that can be processed by each interface per second and smaller packets reach this threshold quicker. When i checked the documentation for the HP switch which has 24ports it says the switch can process 95.2M packets per second so i am wondering why it starts dropping packets after about 2.9M packets

[Joseph W. Doherty]

Yes, that's not all the unusual. As packet size decreases, the PPS rate increases, to sustain the same transmission rate.

You might want to read: https://www.cisco.com/c/en/us/about/security-center/network-performance-metrics.html

[soginy]

thank you for your response i understand that as packet size decreases, the PPS rate increases, to sustain the same transmission rate. is this constant for every switch ? Does this mean that the  maximum frame rate at 1Gb/s is 1,488,096 f/s for every switch. The HP switch i carried the test out on states up to 95.2 Mpps (64-byte packets) in it documentation. its is a 24port gigabit switch so i was expecting it to be able to process about 4Mpps per port based on the documentation

[Joseph W. Doherty · Accepted solution]

" is this constant for every switch ? "

Yes and no. 1.488 Mpps is actually the PPS on the wire, at gig line-speed, for minimal packet sizes. That's what a switch needs to provide if it's to support wire-rate.

"95.2 Mpps (64-byte packets) in it documentation."

Ok, so that means that switch has the capacity for 95.2 / 1.488 = (about) 64 Gbps.

24 ports, needs (about) 36 Mpps. So that switch has "extra" capacity. (Does is support uplink ports, especially 10g?)

[soginy]

yes it does i am using the Aruba 3810M 24G 1-slot Switch (JL071A)
it supports: 
1 open module slot
Supports a maximum of 4 SFP+ ports or 1 40GbE ports, with optional module or 4 Smart Rate ports

[Joseph W. Doherty · Accepted solution]

Ok so 40 Gbps and 24 Gbps would need 95.2 Mpps, which is what you note that device supports. I.e. it's an all port wire-speed capable switch.

However, there's more to how well a switch performs, or not, than its Mpps capacity. A couple of the "biggies" are, a non-blocking architecture and buffer resources and its buffer architecture.