A few questions regarding tier relocation (for all questions, assume auto/continuous/8 hr cycle):
1. What happens when tier relocation does not complete within the cycle time?
a. Are the balance of all pages scheduled for relocation done in the "catch-up" cycle?
b. If not, what happened to the balance of the queued pages?
c. What is the impact of tier relocation that does not complete?
2. In HTnM PD_PLR, is Progress % an accurate indication of the number of pages scheduled to be moved, such that pages_to_be_moved = moved_pages/Progress_%?
3. When tier relocation completes, why does the following happen?
I see the relocation at 00:00 completing, then immediately starting another large relocation, etc. The relocation plan appears to be broken into smaller pieces, such that cycle 1 is actually 6 cycles of relocation, where demotions and promotions are done independently. I know this is a small sample, but I see the same pattern throughout the logs.
4. Given that a 250 TB HDT pool has ~6,000,000 pages and at ~3 TB of relocation/day (the rate I tend to observe on a VSP), so in theory every page could be moved in 80 days (I know this is overly simplistic, but in theory true), with continuous mode, why does tier relocation demand not seem to slow down, overtime, as the pool reaches a somewhat steady state?
a. My expectation is that, once LDEV creation has completed, and the pool achieves a relatively stable state of slow growth, tier relocation queue would tend shrink because new page demand shrinks, "continuous" data becomes mature and the "grey" nature of the boundary calculations for tier residency. Obviously, a dynamic environment would prevent such stability from occurring, but are there any other reasons, in theory, why the queue would not shrink?
Message was edited by: Bryan Mott Edited for clarity