I couldn't find a "methodology" explanation about switch reviews on this site. How are they tested? In them, how is the "cumulative throughput" test performed? The only way I know of to test throughput is to have a host on one port send data to a host on another port (and ensuring that every host used is the same device model), but I don't see how that would work with an odd number of ports, or where 1 port is massively faster than the others like a single 10Gb port.
And the numbers given in the reviews appear to be uni-directional, so how would you be testing 5 ports at a time since one of them wouldn't have a matching source/destination? There is rarely any mention made about whether the switches are non-blocking, and when there is, I'm not sure it's correct. For instance the TP-Link TL-SH183 is said to be non-blocking, but the test results show only about 125Gbps of throughput. That only means that in each pair, one port could send at full speed while the other could receive at the same speed. Being truly non-blocking means that the full speed of every port can be utilized, in BOTH directions at the same time on every port, and the backplane/fabric/chipset has the processing capacity to handle that much traffic, but that would require proving that the total capacity is roughly 280Gbps for that switch. Gigabit switches have had actual non-blocking performance capability for a long time, but 2.5Gb switches that are trying to be faster while not costing much more don't seem to have reached that point, and it is at least something that should be mentioned. (Most use cases won't NEED that much throughput, but expecting it and not getting it would be a problem.)
Manufacturers of low and mid-tier switches like to list "switching capacity" in specs as the total bi-directional speeds of all the ports, even though the actual throughput the chipset is capable of is often only half that (and often not listed). If the only way you can avoid blocking is by limiting how much bandwidth your hosts try to transfer, then the switch isn't actually non-blocking. It would be helpful if the reviews actually tested bi-directional throughput to show whether the backplane/fabric/chipset can actually fully utilize every port at the same time, and to verify that the manufacturer claims are correct. (The claimed throughput on the ASUSTOR ASW209X for example is 44.64Gbps, but they seem to be including the inter-chip link speed in the total which isn't valid.)
And the numbers given in the reviews appear to be uni-directional, so how would you be testing 5 ports at a time since one of them wouldn't have a matching source/destination? There is rarely any mention made about whether the switches are non-blocking, and when there is, I'm not sure it's correct. For instance the TP-Link TL-SH183 is said to be non-blocking, but the test results show only about 125Gbps of throughput. That only means that in each pair, one port could send at full speed while the other could receive at the same speed. Being truly non-blocking means that the full speed of every port can be utilized, in BOTH directions at the same time on every port, and the backplane/fabric/chipset has the processing capacity to handle that much traffic, but that would require proving that the total capacity is roughly 280Gbps for that switch. Gigabit switches have had actual non-blocking performance capability for a long time, but 2.5Gb switches that are trying to be faster while not costing much more don't seem to have reached that point, and it is at least something that should be mentioned. (Most use cases won't NEED that much throughput, but expecting it and not getting it would be a problem.)
Manufacturers of low and mid-tier switches like to list "switching capacity" in specs as the total bi-directional speeds of all the ports, even though the actual throughput the chipset is capable of is often only half that (and often not listed). If the only way you can avoid blocking is by limiting how much bandwidth your hosts try to transfer, then the switch isn't actually non-blocking. It would be helpful if the reviews actually tested bi-directional throughput to show whether the backplane/fabric/chipset can actually fully utilize every port at the same time, and to verify that the manufacturer claims are correct. (The claimed throughput on the ASUSTOR ASW209X for example is 44.64Gbps, but they seem to be including the inter-chip link speed in the total which isn't valid.)
