Road Thrashing Situation 2
Figure 6.3 shows a variation on Scenario 1 in which the Fibre Channel fabric was designed according to VMware best practices. This by itself wouldn’t result in path thrashing. However, the designer decided to customize each host so that they have different preferred paths to LUN 1. These preferred path settings are represented by the interrupted lines (a path from Host A and another path from Host B). The expected behavior in this configuration is that as long as the defined preferred path to LUN 1 is available, the host insists on sending I/O via that path. As a result, Host A attempts to send its I/O to LUN 1 via SPA, and Host B sends it I/O via SPB. However, LUN 1 is owned by SPA and attempts to send I/O via SPB, resulting in a check condition with the sense key ILLEGAL_Demand (more on this in Chapter 7). Host B insists on sending the I/O via its preferred path. So, it sends a START_Unit or a TRESPASS command to the array. As a result, the array transfers LUN 1 ownership to SPB. Now Host A gets really upset and tells the array to transfer the LUN back to SPA, using the START_Product or TRESPASS commands. The array complies, and the tug-of-war begins!
Preventing Highway Thrashing
These advice prompted VMware to make the VMW_PSP_MRU plug-in for fool around with with active/passive arrays. For the old launches, in advance of ESX cuatro.0, so it was previously an insurance plan means for each LUN. Within the 4.0 and later, and 6.0 and you can 6.5, MRU are a beneficial PSA connect-when you look at the. (We the latest PSP construction options during the Chapter 7.) Which have MRU, the latest machine sugar babies Vancouver delivers the I/O towards the most recently put highway. If the LUN moves to another SP, the fresh We/O is distributed with the new path to that SP as an alternative to be taken to SP which was the earlier manager. Observe that MRU ignores the most popular highway means.
ALUA-capable arrays that provide AO AAS for TPGs on the owner SP and ANO AAS for TPGs on the non-owner SP allow I/O to the given LUN with high priority via the AO TPGs and, conversely, lower priority via the ANO TPGs. This means that the latter does not return a check condition with sense key ILLEGAL_Request if I/O to the LUN is sent through it. This means that using VMW_PSP_FIXED with these arrays can result in a lighter version of path thrashing. In this case, I/O does not fail to be sent to the ANO TPGs if that is the preferred path. However, the I/O performance is much lower compared to using the AO TPGs. If more hosts are using the AO TPGs as the preferred path, the LUN ownership stays on the original SP that owns it. As a result, the ANO TPGs are not transitioned to AO for the offending host.
To accommodate this example, VMware put an alternate function for use which have ALUA devices; not, this is not laid out from the ALUA spec. This feature is referred to as ALUA followover.
ALUA followover simply implies that if the host finds a TPG AAS change so it failed to end in alone, it will not just be sure to revert the alteration no matter if they only has the means to access TPGs which might be ANO. Effortlessly, it prevents the fresh new machines from fighting for TPG AAS and, as an alternative, it stick to the TPG AAS of your variety. Numbers six.4 and you will six.5 train ALUA followover telecommunications that have TPG AAS.
Profile 6.4 shows a systematic stores diagram in which the switch fabrics was basically got rid of to explain the new drawing . Right here, TPG ID 1 ‘s the AO toward Salon, and you may each other servers publish the brand new I/O to that TPG. TPG ID dos is ANO, and i/O isn’t delivered to they. Such TPGs is actually designed having ALUA Specific function.