Additional layout policies + more flexibility in buffer flushing

2 files changed, 131 insertions, 0 deletions

diff --git a/include/framework/reconstruction/FixedShardCountPolicy.h b/include/framework/reconstruction/FixedShardCountPolicy.h
new file mode 100644
index 0000000..ec8e4e6
--- /dev/null
+++ b/include/framework/reconstruction/FixedShardCountPolicy.h
@@ -0,0 +1,66 @@
+/*
+ * include/framework/reconstruction/FixedShardCountPolicy.h
+ *
+ * Copyright (C) 2023-2024 Douglas B. Rumbaugh <drumbaugh@psu.edu>
+ *                         Dong Xie <dongx@psu.edu>
+ *
+ * Distributed under the Modified BSD License.
+ *
+ */
+#pragma once
+#include <cmath>
+
+#include "framework/reconstruction/ReconstructionPolicy.h"
+#include "framework/scheduling/Epoch.h"
+#include "util/types.h"
+
+namespace de {
+template <ShardInterface ShardType, QueryInterface<ShardType> QueryType>
+class FixedShardCountPolicy : public ReconstructionPolicy<ShardType, QueryType> {
+  typedef std::vector<std::shared_ptr<InternalLevel<ShardType, QueryType>>>
+      LevelVector;
+
+public:
+  FixedShardCountPolicy(size_t buffer_size, size_t shard_count, size_t max_record_count) 
+  : m_buffer_size(buffer_size), m_shard_count(shard_count), m_max_reccnt(max_record_count) {}
+
+  ReconstructionVector
+  get_reconstruction_tasks(const Epoch<ShardType, QueryType> *epoch,
+                           size_t incoming_reccnt) const override {
+    ReconstructionVector reconstructions;
+    return reconstructions;
+
+  }
+
+  ReconstructionTask
+  get_flush_task(const Epoch<ShardType, QueryType> *epoch) const override {
+
+    auto levels = epoch->get_structure()->get_level_vector();
+
+    if (levels.size() == 0) {
+      return ReconstructionTask{
+          {{buffer_shid}}, 0, m_buffer_size, ReconstructionType::Append};
+    }
+
+    ShardID last_shid = {0, (shard_index) (levels[0]->get_shard_count() - 1)};
+
+    if (levels[0]->get_shard(last_shid.shard_idx)->get_record_count() + m_buffer_size <= capacity()) {
+        return ReconstructionTask{
+          {{buffer_shid, last_shid}}, 0, m_buffer_size, ReconstructionType::Merge};
+    } else {
+        return ReconstructionTask{
+          {{buffer_shid}}, 0, m_buffer_size, ReconstructionType::Append};
+    }
+  }
+
+private:
+  inline size_t capacity() const {
+    double bps = (double) m_max_reccnt / (double) m_buffer_size / m_shard_count;
+    return ceil(bps) * m_buffer_size;
+  }
+
+  size_t m_buffer_size;
+  size_t m_shard_count;
+  size_t m_max_reccnt;
+};
+} // namespace de
diff --git a/include/framework/reconstruction/FloodL0Policy.h b/include/framework/reconstruction/FloodL0Policy.h
new file mode 100644
index 0000000..da4c297
--- /dev/null
+++ b/include/framework/reconstruction/FloodL0Policy.h
@@ -0,0 +1,65 @@
+/*
+ * include/framework/reconstruction/FloodL0Policy.h
+ *
+ * Copyright (C) 2023-2024 Douglas B. Rumbaugh <drumbaugh@psu.edu>
+ *                         Dong Xie <dongx@psu.edu>
+ *
+ * Distributed under the Modified BSD License.
+ *
+ */
+#pragma once
+#include <cmath>
+
+#include "framework/reconstruction/ReconstructionPolicy.h"
+#include "framework/scheduling/Epoch.h"
+#include "util/types.h"
+
+namespace de {
+template <ShardInterface ShardType, QueryInterface<ShardType> QueryType>
+class FloodL0Policy : public ReconstructionPolicy<ShardType, QueryType> {
+  typedef std::vector<std::shared_ptr<InternalLevel<ShardType, QueryType>>>
+      LevelVector;
+
+public:
+  FloodL0Policy(size_t buffer_size) : m_buffer_size(buffer_size) {}
+
+  ReconstructionVector
+  get_reconstruction_tasks(const Epoch<ShardType, QueryType> *epoch,
+                           size_t incoming_reccnt) const override {
+
+    ReconstructionVector reconstructions;
+    return reconstructions;
+
+  }
+
+  ReconstructionTask
+  get_flush_task(const Epoch<ShardType, QueryType> *epoch) const override {
+    return ReconstructionTask{
+        {{buffer_shid}}, 0, m_buffer_size, ReconstructionType::Append};
+  }
+
+private:
+  level_index find_reconstruction_target(LevelVector &levels) const {
+    level_index target_level = invalid_level_idx;
+    size_t incoming_records = m_buffer_size;
+
+    for (level_index i = 0; i < (level_index)levels.size(); i++) {
+      if (levels[i]->get_record_count() + incoming_records < capacity(i)) {
+        target_level = i;
+        break;
+      }
+
+      incoming_records = levels[i]->get_record_count();
+    }
+
+    return target_level;
+  }
+
+  inline size_t capacity(level_index level) const {
+    return m_buffer_size * pow(m_scale_factor, level + 1);
+  }
+
+  size_t m_scale_factor;
+  size_t m_buffer_size;
+};
+} // namespace de