Re: [PoC] Improve dead tuple storage for lazy vacuum - Mailing list pgsql-hackers

> One issue with this patch: The "fanout" member is a uint8, so it can't hold 256 for the largest node kind. That's not an issue in practice, since we never need to grow it, and we only compare that value with the count in an Assert(), so I just set it to zero. That does break an invariant, so it's not great. We could use 2 bytes to be strictly correct in all cases, but that limits what we can do with the smallest node kind.

Thinking about this part, there's an easy resolution -- use a different macro for fixed- and variable-sized node kinds to determine if there is a free slot.

Also, I wanted to share some results of adjusting the boundary between the two smallest node kinds. In the hackish attached patch, I modified the fixed height search benchmark to search a small (within L1 cache) tree thousands of times. For the first set I modified node4's maximum fanout and filled it up. For the second, I set node4's fanout to 1, which causes 2+ to spill to node32 (actually the partially-filled node15 size class as demoed earlier).

node4:

NOTICE:  num_keys = 16, height = 3, n4 = 15, n15 = 0, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      2 |    16 |            16520 |          0 |            3

NOTICE:  num_keys = 81, height = 3, n4 = 40, n15 = 0, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      3 |    81 |            16456 |          0 |           17

NOTICE:  num_keys = 256, height = 3, n4 = 85, n15 = 0, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      4 |   256 |            16456 |          0 |           89

NOTICE:  num_keys = 625, height = 3, n4 = 156, n15 = 0, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      5 |   625 |            16488 |          0 |          327


node32:

NOTICE:  num_keys = 16, height = 3, n4 = 0, n15 = 15, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      2 |    16 |            16488 |          0 |            5
(1 row)

NOTICE:  num_keys = 81, height = 3, n4 = 0, n15 = 40, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      3 |    81 |            16520 |          0 |           28

NOTICE:  num_keys = 256, height = 3, n4 = 0, n15 = 85, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      4 |   256 |            16408 |          0 |           79

NOTICE:  num_keys = 625, height = 3, n4 = 0, n15 = 156, n32 = 0, n128 = 0, n256 = 0
 fanout | nkeys | rt_mem_allocated | rt_load_ms | rt_search_ms
--------+-------+------------------+------------+--------------
      5 |   625 |            24616 |          0 |          199

In this test, node32 seems slightly faster than node4 with 4 elements, at the cost of more memory. 

Assuming the smallest node is fixed size (i.e. fanout/capacity member not part of the common set, so only part of variable-sized nodes), 3 has a nice property: no wasted padding space:

node4: 5 + 4+(7) + 4*8 = 48 bytes
node3: 5 + 3     + 3*8 = 32

--
John Naylor
EDB: http://www.enterprisedb.com