TrieDB: use im::OrdMap to reduce memory consumption

[kvinwang]

The im::HashMap has a really heavy memory overhead. Let's use im::OrdMap to reduce memory usage.

Drawback

im::OrdMap is a bit slower than im::HashMap. It took about 57 minutes to sync from block height 0 to 900000 vs 52 minutes for im::HashMap.

[jasl]

LGTM, have you compare saving checkpoint of these data structures?

[kvinwang]

LGTM, have you compare saving checkpoint of these data structures?

No, I haven't. There is no significant difference in the early blocks. In order to make a meaningful comparison, I think it needs to sync to recent blocks which make big-size checkpoints.

@jasl Could you build a test version and ask @doyleguo to test it?

[jasl]

LGTM, have you compare saving checkpoint of these data structures?

No, I haven't. There is no significant difference in the early blocks. In order to make a meaningful comparison, I think it needs to sync to recent blocks which make big-size checkpoints.

@jasl Could you build a test version and ask @doyleguo to test it?

No Problem, this is PRuntime V2 right? BTW, does this need to back port to v0 to replace std::Hash one?

[kvinwang]

this is PRuntime V2 right? BTW, does this need to back port to v0 to replace std::Hash one?

Yep, this is V2. For v0 we already replaced std::HashMap with std::BTreeMap in #1121. It doesn't need to backport this PR.

im::OrdMap is almost the same as std::BTreeMap on memory consumption.

[h4x3rotab]

LGTM.