Consider assorted data structure/storage optimizations

[alnoki]

• Simple red black tree for price levels in a single storage item
• Orders within a price level as ring-buffer based doubly linked list
  • Ring buffer so can know next allocation spot without having to always reshuffle (e.g. like in a vector insert)
    • The allocated vector never changes, just the next and last fields
    • 2d ring buffer could have max elems in a storage node, and have each node be doubly linked list yet again (like BigVector)
  • Hence for O(1) lookup need pointer to table ring buffer and vector ring buffer
• Except if static space for price levels then need to evict entire level at once
• Could use btree on its own for all orders, since maker address already 8 bytes and other data probably 30 or so
• Could use b+ tree
  • Each order has a b+ tree key, which is price concatenated with a boundary flag
  • Basically if you have an order at price 5 and price 6 and need to insert at back of queue for price 5, look at the boundary flag on existing price 5 order (e.g. 000000), and increment by 1, so new order key is5 < 64 | 1 (this is like the dynamic pointer used for AVL queue) and thus goes between the two existing orders in total order search
  • Then need to store B+ access key under user table