checkpoint only (shadow page technique version fixed)

README.md

@@ -55,6 +55,7 @@ $ ./build.sh --origin
 | innodb_nvdimm_buffer_pool_instances | The number of regions that the NVDIMM cache is divided into. The default value is 1. |
 | innodb_nvdimm_pc_threshold_pct      | Wakeup the NVDIMM page cleaner when this % of free pages remaining. The default value is 5. |
 | innodb_nvdimm_home_dir				      | NVDIMM-aware files resident directory |
+| innodb_use_nvdimm_dwb				        | Specifies whether to use NVDIMM dwb. **true** or **false**. |
 
 For example:
 

build.sh

@@ -29,15 +29,15 @@ elif [ "$1" = "--nc-monitor" ]; then
     # Cache hot LB pages with mtr-logging/monitoring enabled
     BUILD_FLAGS="-DUNIV_NVDIMM_CACHE -DUNIV_LOG_HEADER -DUNIV_FLUSH_MONITOR"
 else
-    # Cache hot LB pages (default)
-    BUILD_FLAGS="-DUNIV_NVDIMM_CACHE -DUNIV_LOG_HEADER"
+    # Cache NVDIMM pages in TPC-C workloads
+    BUILD_FLAGS="-DUNIV_NVDIMM_CACHE"
 fi
 
 echo "Start build using $BUILD_FLAGS"
 
-cd $BASE_DIR
+cd $BUILD_DIR
 
-cmake -DWITH_DEBUG=0 -DCMAKE_C_FLAGS="$BUILD_FLAGS" -DCMAKE_CXX_FLAGS="$BUILD_FLAGS" \
+cmake .. -DWITH_DEBUG=0 -DCMAKE_C_FLAGS="$BUILD_FLAGS" -DCMAKE_CXX_FLAGS="$BUILD_FLAGS" \
 -DDOWNLOAD_BOOST=ON -DWITH_BOOST=$BASE_DIR/boost -DENABLED_LOCAL_INFILE=1 \
 -DCMAKE_INSTALL_PREFIX=$BUILD_DIR
 

my.cnf

@@ -14,26 +14,31 @@ prompt=\u:\d>\_
 
 # This was formally known as [safe_mysqld]. Both versions are currently parsed.
 [mysqld_safe]
+user = root
 socket          = /tmp/mysql.sock
 #nice           = 0
 
 [mysqld]
 #
 # * Basic Settings
 #
+user = root
 default-storage-engine = innodb
 skip-grant-tables
-pid-file        = /home/mijin/test_data/mysql.pid
+pid-file        = /home/vldb/test_data/mysql.pid
 socket          = /tmp/mysql.sock
 port            = 3306
-datadir         = /home/mijin/test_data/
-log-error	= /home/mijin/test_data/mysql_error_nvdimm.log
+datadir         = /home/vldb/test_data/
+log-error	= /home/vldb/test_data/mysql_error_nvdimm.log
 
 ##################################################
 # Need to Modify
 ##################################################
 #Log group path (iblog0, iblog1)
-innodb_log_group_home_dir=/home/mijin/test_log/
+innodb_log_group_home_dir=/home/vldb/test_log/
+#innodb_log_group_home_dir=/mnt/pmemdir/test_log/
+#innodb_undo_directory=/mnt/pmemdir/
+
 
 #innodb page size
 innodb_page_size=4KB
@@ -59,21 +64,24 @@ innodb_use_nvdimm_buffer=true
 innodb_nvdimm_buffer_pool_size=1G
 innodb_nvdimm_buffer_pool_instances=1
 innodb_nvdimm_pc_threshold_pct=15
-innodb_nvdimm_home_dir=/mnt/ramdisk
-#innodb_nvdimm_home_dir=/mnt/pmem
+innodb_nvdimm_home_dir=/mnt/pmemdir
 
 #transaction log settings
-innodb_log_file_size=2G
+innodb_log_file_size=500M
 innodb_log_files_in_group=3
 
 # 0:every 1 seconds, 1:fsync on commits, 2:writes on commits
-innodb_flush_log_at_trx_commit=0
+innodb_flush_log_at_trx_commit=1
 innodb_log_buffer_size=32M
 innodb_flush_neighbors=0
+#innodb_force_recovery = 1
 
 #doublewrite and flush method
 innodb_doublewrite=ON
 innodb_flush_method=O_DIRECT
+# dwb NVM version
+innodb_use_nvdimm_dwb=false
+
 
 #AIO control
 innodb_use_native_aio=true
@@ -95,4 +103,3 @@ open_files_limit = 24000
 #performance-schema-instrument='wait/synch/rwlock/innodb/%=ON'
 #innodb_status_output = ON
 #innodb_status_output_locks = ON
-

storage/innobase/btr/btr0cur.cc

@@ -3924,23 +3924,19 @@ btr_cur_update_in_place(
 	}
 
 #ifdef UNIV_NVDIMM_CACHE
+    // (jhpark): add REDO log for NC pages 
+    btr_cur_update_in_place_log(flags, rec, index, update,
+                        trx_id, roll_ptr, mtr);
+    /*
     nvm_block = btr_cur_get_block(cursor);
     nvm_bpage = &(nvm_block->page);
-
     if (nvm_bpage->cached_in_nvdimm) {
-        // skip generating REDO logs for NVM-resident pages
-				// write NC page on NVDIMM
-				//pm_mmap_buf_write(nvm_bpage->size.physical(), (void*) ((buf_block_t*) nvm_bpage)->frame);
-
-				// persist records
-				ulint cur_rec_size = rec_offs_size(offsets);
-                pm_mmap_mtrlogbuf_commit(nvm_block->frame, UNIV_PAGE_SIZE, nvm_bpage->id.space(), nvm_bpage->id.page_no());
-
-				//pm_mmap_mtrlogbuf_commit(rec, cur_rec_size, nvm_bpage->id.space(), nvm_bpage->id.page_no());
+
     } else {
         btr_cur_update_in_place_log(flags, rec, index, update,
                         trx_id, roll_ptr, mtr);
     }
+    */
 #else
 	btr_cur_update_in_place_log(flags, rec, index, update,
 				    trx_id, roll_ptr, mtr);
@@ -4937,6 +4933,7 @@ btr_cur_del_mark_set_clust_rec(
 	row_upd_rec_sys_fields(rec, page_zip, index, offsets, trx, roll_ptr);
 
 #ifdef UNIV_NVDIMM_CACHE
+  /*
     if (is_nvm_page) {
     	// skip generating REDO logs for nvm-page
 			// Instead, write commit log in mtr log
@@ -4950,6 +4947,12 @@ btr_cur_del_mark_set_clust_rec(
         btr_cur_del_mark_set_clust_rec_log(rec, index, trx->id,
 					    roll_ptr, mtr);
     }
+   */
+
+  // (jhpark): add REDO log for NC pages
+  btr_cur_del_mark_set_clust_rec_log(rec, index, trx->id,
+      roll_ptr, mtr);
+
 #else
 	btr_cur_del_mark_set_clust_rec_log(rec, index, trx->id,
 					   roll_ptr, mtr);

storage/innobase/buf/buf0buf.cc

@@ -406,6 +406,74 @@ buf_pool_register_chunk(
 		chunk->blocks->frame, chunk));
 }
 
+// HOT DEBUG //
+lsn_t
+nvdimm_buf_pool_get_oldest_modification(void)
+/*==================================*/
+{
+	lsn_t		lsn = 0;
+	lsn_t		oldest_lsn = 0;
+  lsn_t   page_lsn = 0;
+
+	/* When we traverse all the flush lists we don't want another
+	thread to add a dirty page to any flush list. */
+	log_flush_order_mutex_enter();
+
+	for (ulint i = srv_buf_pool_instances; i < srv_buf_pool_instances+1; i++) {
+		buf_pool_t*	buf_pool;
+
+		buf_pool = buf_pool_from_array(i);
+
+		buf_flush_list_mutex_enter(buf_pool);
+
+		buf_page_t*	bpage;
+
+		/* We don't let log-checkpoint halt because pages from system
+		temporary are not yet flushed to the disk. Anyway, object
+		residing in system temporary doesn't generate REDO logging. */
+		for (bpage = UT_LIST_GET_LAST(buf_pool->flush_list);
+		     bpage != NULL
+			&& fsp_is_system_temporary(bpage->id.space());
+		     bpage = UT_LIST_GET_PREV(list, bpage)) {
+			/* Do nothing. */
+		}
+
+		if (bpage != NULL) {
+			ut_ad(bpage->in_flush_list);
+
+      // HOT DEBUG //
+      // check page lsn of current NC pages
+      /*
+      buf_block_t *block;
+      block = (buf_block_t*)bpage;
+      uint64_t cur_lsn_page = mach_read_from_8(block->frame + FIL_PAGE_LSN);
+      if (cur_lsn_page !=0 
+          && cur_lsn_page < pmem_lsn) {
+        lsn = pmem_lsn;
+      } else {
+        lsn = cur_lsn_page;
+      }
+      */
+      lsn = bpage->oldest_modification;
+		}
+
+		buf_flush_list_mutex_exit(buf_pool);
+
+		if (!oldest_lsn || oldest_lsn > lsn) {
+			oldest_lsn = lsn;
+		}
+	}
+
+	log_flush_order_mutex_exit();
+
+	/* The returned answer may be out of date: the flush_list can
+	change after the mutex has been released. */
+
+	return(oldest_lsn);
+}
+
+
+
 /********************************************************************//**
 Gets the smallest oldest_modification lsn for any page in the pool. Returns
 zero if all modified pages have been flushed to disk.
@@ -421,7 +489,7 @@ buf_pool_get_oldest_modification(void)
 	thread to add a dirty page to any flush list. */
 	log_flush_order_mutex_enter();
 
-	for (ulint i = 0; i < srv_buf_pool_instances; i++) {
+  for (ulint i = 0; i < srv_buf_pool_instances; i++) {
 		buf_pool_t*	buf_pool;
 
 		buf_pool = buf_pool_from_array(i);
@@ -6331,7 +6399,7 @@ buf_page_io_complete(
 Asserts that all file pages in the buffer are in a replaceable state.
 @return TRUE */
 static
-ibool
+          ibool
 buf_all_freed_instance(
 /*===================*/
 	buf_pool_t*	buf_pool)	/*!< in: buffer pool instancce */