performance: malloc on insert

[roblatham00]

the mercury folks learned this the hard way: malloc on every operation increases latency. We talked about this for a while in the group meeting.

[halimamer]

Efficient alloc/free of queue and list elements is in our to-do list. We will proceed in two steps:

1. Implement a scalable allocator for internal allocations
2. Extend the API to offload memory management to Izem.

Currently lists and queues require potentially double alloc/free operations: one set by the user who might allocate the data and pass the pointer to Izem; and another set for the internal list/queue element.

1. will ensure fast internal memory management (goals: lock-free and cache-friendly). 2) Will avoid double alloc/free operations, and the single alloc/free set of operations will be expected to be well implemented.