We are using 2920 series switches dedicated to iSCSI traffic (we use the 10Gb SFP+ uplinks for the traffic) e.g. J9728A As the switches are dedicated to iSCSI and because there is only one IP subnet per physical switch, there is no need for vLANs. So all traffic is running over what is – in HPE parlance – the Default VLAN (ID 1). There are also no QoS rules configured because all of the iSCSI traffic is (in networking terms) equally urgent. Although the packet buffer on these switches is not huge when compared with some of the more upmarket alternatives, it is OK and seems decent enough for our iSCSI requirements. However, as we are about to expand our storage using a second array, I wanted to take the opportunity to check about the QoS Queues and see if it is possible to apply any tweaks for maximum potential performance. My understanding is that these 2920 switches can be configured to use 2, 4 or 8 QoS queues. And that the factory-default setting is 4. First Question: since the buffer size will remain constant, will reducing the number of queues increase the maximum depth of each queue? If I reconfigure the switches to use 2 queues instead of 4, will the available buffer resources be divided into halves instead of quarters? Or does it simply not work that way? I like the idea of deeper queues because presumably, all other things being equal, we are less likely to see packets dropped. If it does work this way, then I have a Second Question: will modifying the number of queues actually make any difference when there are no QoS rules in place? Are these queues utilised with the switch in its factory-default state vis-à-vis QoS? If it doesn’t work this way, can anyone offer up some tips as to how it does work, and in what circumstances one would want to configure a higher or lower quantity of queues?
No comments:
Post a Comment