Improve large query optimizer pass

Former-commit-id: 1f0c283ec4dc5560c14d4e17d76719d38dfc2280

CHANGELOG.md

@@ -21,6 +21,7 @@ Please mark all change in change log and use the ticket from JIRA.
 - \#118 - Using shared_ptr instead of weak_ptr to avoid performance loss
 - \#122 - Add unique id for Job
 - \#130 - Set task state MOVED after resource copy it completed
+- \#149 - Improve large query optimizer pass
 
 ## Task
 

core/src/db/engine/ExecutionEngineImpl.cpp

@@ -258,7 +258,7 @@ Status
 ExecutionEngineImpl::CopyToGpu(uint64_t device_id, bool hybrid) {
     if (hybrid) {
         const std::string key = location_ + ".quantizer";
-        std::vector<uint64_t> gpus = scheduler::get_gpu_pool();
+        std::vector<uint64_t> gpus{device_id};
 
         const int64_t NOT_FOUND = -1;
         int64_t device_id = NOT_FOUND;

core/src/scheduler/optimizer/LargeSQ8HPass.cpp

@@ -55,16 +55,18 @@ LargeSQ8HPass::Run(const TaskPtr& task) {
     }
 
     std::vector<uint64_t> gpus = scheduler::get_gpu_pool();
-    std::vector<int64_t> all_free_mem;
-    for (auto& gpu : gpus) {
-        auto cache = cache::GpuCacheMgr::GetInstance(gpu);
-        auto free_mem = cache->CacheCapacity() - cache->CacheUsage();
-        all_free_mem.push_back(free_mem);
-    }
-
-    auto max_e = std::max_element(all_free_mem.begin(), all_free_mem.end());
-    auto best_index = std::distance(all_free_mem.begin(), max_e);
-    auto best_device_id = gpus[best_index];
+    //    std::vector<int64_t> all_free_mem;
+    //    for (auto& gpu : gpus) {
+    //        auto cache = cache::GpuCacheMgr::GetInstance(gpu);
+    //        auto free_mem = cache->CacheCapacity() - cache->CacheUsage();
+    //        all_free_mem.push_back(free_mem);
+    //    }
+    //
+    //    auto max_e = std::max_element(all_free_mem.begin(), all_free_mem.end());
+    //    auto best_index = std::distance(all_free_mem.begin(), max_e);
+    //    auto best_device_id = gpus[best_index];
+    auto best_device_id = count_ % gpus.size();
+    count_++;
 
     ResourcePtr res_ptr = ResMgrInst::GetInstance()->GetResource(ResourceType::GPU, best_device_id);
     if (not res_ptr) {

core/src/scheduler/optimizer/LargeSQ8HPass.h

@@ -43,6 +43,7 @@ class LargeSQ8HPass : public Pass {
 
  private:
     int32_t threshold_ = std::numeric_limits<int32_t>::max();
+    int64_t count_ = 0;
 };
 
 using LargeSQ8HPassPtr = std::shared_ptr<LargeSQ8HPass>;