Pre-allocated free space for row updating (like PCTFREE)

Hi all,

I've done a quick hack to implement PCTFREE on PostgreSQL.

As you know, it's inspired by Oracle's PCTFREE.

http://www.csee.umbc.edu/help/oracle8/server.815/a67772/schema.htm#990
http://www.comp.hkbu.edu.hk/docs/o/oracle10g/server.101/b10743/cncpt031.gif

Pre-allocated space for each block(page) can improve heap_update() performance,
because heap_update() looks for the free space in same block
to insert new row.

According to my experiments, pgbench score was improved 10% or more
with 1024 bytes free space.

Any comments? Is this idea good, or not?

Thanks.
--
NAGAYASU Satoshi <nagayasus@nttdata.co.jp>
diff -rc postgresql-8.0.0.orig/src/backend/access/heap/heapam.c
postgresql-8.0.0.pctfree/src/backend/access/heap/heapam.c
*** postgresql-8.0.0.orig/src/backend/access/heap/heapam.c    2005-01-01 06:59:16.000000000 +0900
--- postgresql-8.0.0.pctfree/src/backend/access/heap/heapam.c    2005-08-20 23:20:45.017901208 +0900
***************
*** 1151,1157 ****
          heap_tuple_toast_attrs(relation, tup, NULL);

      /* Find buffer to insert this tuple into */
!     buffer = RelationGetBufferForTuple(relation, tup->t_len, InvalidBuffer);

      /* NO EREPORT(ERROR) from here till changes are logged */
      START_CRIT_SECTION();
--- 1151,1160 ----
          heap_tuple_toast_attrs(relation, tup, NULL);

      /* Find buffer to insert this tuple into */
!     buffer = RelationGetBufferForTuple(relation,
!                                        tup->t_len,
!                                        InvalidBuffer,
!                                        true);

      /* NO EREPORT(ERROR) from here till changes are logged */
      START_CRIT_SECTION();
***************
*** 1671,1678 ****
          if (newtupsize > pagefree)
          {
              /* Assume there's no chance to put newtup on same page. */
!             newbuf = RelationGetBufferForTuple(relation, newtup->t_len,
!                                                buffer);
          }
          else
          {
--- 1674,1683 ----
          if (newtupsize > pagefree)
          {
              /* Assume there's no chance to put newtup on same page. */
!             newbuf = RelationGetBufferForTuple(relation,
!                                                newtup->t_len,
!                                                buffer,
!                                                false);
          }
          else
          {
***************
*** 1688,1695 ****
                   * should seldom be taken.
                   */
                  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
!                 newbuf = RelationGetBufferForTuple(relation, newtup->t_len,
!                                                    buffer);
              }
              else
              {
--- 1693,1702 ----
                   * should seldom be taken.
                   */
                  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
!                 newbuf = RelationGetBufferForTuple(relation,
!                                                    newtup->t_len,
!                                                    buffer,
!                                                    false);
              }
              else
              {
diff -rc postgresql-8.0.0.orig/src/backend/access/heap/hio.c postgresql-8.0.0.pctfree/src/backend/access/heap/hio.c
*** postgresql-8.0.0.orig/src/backend/access/heap/hio.c    2005-01-01 06:59:16.000000000 +0900
--- postgresql-8.0.0.pctfree/src/backend/access/heap/hio.c    2005-08-20 23:35:44.986085248 +0900
***************
*** 89,95 ****
   */
  Buffer
  RelationGetBufferForTuple(Relation relation, Size len,
!                           Buffer otherBuffer)
  {
      Buffer        buffer = InvalidBuffer;
      Page        pageHeader;
--- 89,95 ----
   */
  Buffer
  RelationGetBufferForTuple(Relation relation, Size len,
!                           Buffer otherBuffer, bool forInsert)
  {
      Buffer        buffer = InvalidBuffer;
      Page        pageHeader;
***************
*** 136,142 ****
           * We have no cached target page, so ask the FSM for an initial
           * target.
           */
!         targetBlock = GetPageWithFreeSpace(&relation->rd_node, len);

          /*
           * If the FSM knows nothing of the rel, try the last page before
--- 136,142 ----
           * We have no cached target page, so ask the FSM for an initial
           * target.
           */
!         targetBlock = GetPageWithFreeSpace(&relation->rd_node, len, forInsert);

          /*
           * If the FSM knows nothing of the rel, try the last page before
***************
*** 192,198 ****
           */
          pageHeader = (Page) BufferGetPage(buffer);
          pageFreeSpace = PageGetFreeSpace(pageHeader);
!         if (len <= pageFreeSpace)
          {
              /* use this page as future insert target, too */
              relation->rd_targblock = targetBlock;
--- 192,198 ----
           */
          pageHeader = (Page) BufferGetPage(buffer);
          pageFreeSpace = PageGetFreeSpace(pageHeader);
!         if ((forInsert ? (len+1024) : len) <= pageFreeSpace)
          {
              /* use this page as future insert target, too */
              relation->rd_targblock = targetBlock;
***************
*** 221,227 ****
          targetBlock = RecordAndGetPageWithFreeSpace(&relation->rd_node,
                                                      targetBlock,
                                                      pageFreeSpace,
!                                                     len);
      }

      /*
--- 221,228 ----
          targetBlock = RecordAndGetPageWithFreeSpace(&relation->rd_node,
                                                      targetBlock,
                                                      pageFreeSpace,
!                                                     len,
!                                                     forInsert);
      }

      /*
diff -rc postgresql-8.0.0.orig/src/backend/storage/freespace/freespace.c
postgresql-8.0.0.pctfree/src/backend/storage/freespace/freespace.c
*** postgresql-8.0.0.orig/src/backend/storage/freespace/freespace.c    2005-01-01 07:00:54.000000000 +0900
--- postgresql-8.0.0.pctfree/src/backend/storage/freespace/freespace.c    2005-08-20 23:18:21.413732360 +0900
***************
*** 229,235 ****
  static void unlink_fsm_rel_usage(FSMRelation *fsmrel);
  static void link_fsm_rel_storage(FSMRelation *fsmrel);
  static void unlink_fsm_rel_storage(FSMRelation *fsmrel);
! static BlockNumber find_free_space(FSMRelation *fsmrel, Size spaceNeeded);
  static BlockNumber find_index_free_space(FSMRelation *fsmrel);
  static void fsm_record_free_space(FSMRelation *fsmrel, BlockNumber page,
                        Size spaceAvail);
--- 229,237 ----
  static void unlink_fsm_rel_usage(FSMRelation *fsmrel);
  static void link_fsm_rel_storage(FSMRelation *fsmrel);
  static void unlink_fsm_rel_storage(FSMRelation *fsmrel);
! static BlockNumber find_free_space(FSMRelation *fsmrel,
!                                    Size spaceNeeded,
!                                    bool forInsert);
  static BlockNumber find_index_free_space(FSMRelation *fsmrel);
  static void fsm_record_free_space(FSMRelation *fsmrel, BlockNumber page,
                        Size spaceAvail);
***************
*** 359,365 ****
   * extend the relation.
   */
  BlockNumber
! GetPageWithFreeSpace(RelFileNode *rel, Size spaceNeeded)
  {
      FSMRelation *fsmrel;
      BlockNumber freepage;
--- 361,367 ----
   * extend the relation.
   */
  BlockNumber
! GetPageWithFreeSpace(RelFileNode *rel, Size spaceNeeded, bool forInsert)
  {
      FSMRelation *fsmrel;
      BlockNumber freepage;
***************
*** 384,390 ****
          cur_avg += ((int) spaceNeeded - cur_avg) / 32;
          fsmrel->avgRequest = (Size) cur_avg;
      }
!     freepage = find_free_space(fsmrel, spaceNeeded);
      LWLockRelease(FreeSpaceLock);
      return freepage;
  }
--- 386,392 ----
          cur_avg += ((int) spaceNeeded - cur_avg) / 32;
          fsmrel->avgRequest = (Size) cur_avg;
      }
!     freepage = find_free_space(fsmrel, spaceNeeded, forInsert);
      LWLockRelease(FreeSpaceLock);
      return freepage;
  }
***************
*** 399,405 ****
  RecordAndGetPageWithFreeSpace(RelFileNode *rel,
                                BlockNumber oldPage,
                                Size oldSpaceAvail,
!                               Size spaceNeeded)
  {
      FSMRelation *fsmrel;
      BlockNumber freepage;
--- 401,408 ----
  RecordAndGetPageWithFreeSpace(RelFileNode *rel,
                                BlockNumber oldPage,
                                Size oldSpaceAvail,
!                               Size spaceNeeded,
!                               bool forInsert)
  {
      FSMRelation *fsmrel;
      BlockNumber freepage;
***************
*** 429,435 ****
          fsmrel->avgRequest = (Size) cur_avg;
      }
      /* Do the Get */
!     freepage = find_free_space(fsmrel, spaceNeeded);
      LWLockRelease(FreeSpaceLock);
      return freepage;
  }
--- 432,438 ----
          fsmrel->avgRequest = (Size) cur_avg;
      }
      /* Do the Get */
!     freepage = find_free_space(fsmrel, spaceNeeded, forInsert);
      LWLockRelease(FreeSpaceLock);
      return freepage;
  }
***************
*** 1204,1210 ****
   * if no success.
   */
  static BlockNumber
! find_free_space(FSMRelation *fsmrel, Size spaceNeeded)
  {
      FSMPageData *info;
      int            pagesToCheck,    /* outer loop counter */
--- 1207,1213 ----
   * if no success.
   */
  static BlockNumber
! find_free_space(FSMRelation *fsmrel, Size spaceNeeded, bool forInsert)
  {
      FSMPageData *info;
      int            pagesToCheck,    /* outer loop counter */
***************
*** 1225,1231 ****
          Size        spaceAvail = FSMPageGetSpace(page);

          /* Check this page */
!         if (spaceAvail >= spaceNeeded)
          {
              /*
               * Found what we want --- adjust the entry, and update
--- 1228,1234 ----
          Size        spaceAvail = FSMPageGetSpace(page);

          /* Check this page */
!         if (spaceAvail >= (forInsert ? (spaceNeeded + 1024) : spaceNeeded) )
          {
              /*
               * Found what we want --- adjust the entry, and update
diff -rc postgresql-8.0.0.orig/src/include/access/hio.h postgresql-8.0.0.pctfree/src/include/access/hio.h
*** postgresql-8.0.0.orig/src/include/access/hio.h    2005-01-01 07:03:21.000000000 +0900
--- postgresql-8.0.0.pctfree/src/include/access/hio.h    2005-08-20 23:13:20.267513544 +0900
***************
*** 19,24 ****
  extern void RelationPutHeapTuple(Relation relation, Buffer buffer,
                       HeapTuple tuple);
  extern Buffer RelationGetBufferForTuple(Relation relation, Size len,
!                           Buffer otherBuffer);

  #endif   /* HIO_H */
--- 19,24 ----
  extern void RelationPutHeapTuple(Relation relation, Buffer buffer,
                       HeapTuple tuple);
  extern Buffer RelationGetBufferForTuple(Relation relation, Size len,
!                           Buffer otherBuffer, bool forInsert);

  #endif   /* HIO_H */
diff -rc postgresql-8.0.0.orig/src/include/storage/freespace.h postgresql-8.0.0.pctfree/src/include/storage/freespace.h
*** postgresql-8.0.0.orig/src/include/storage/freespace.h    2005-01-01 07:03:42.000000000 +0900
--- postgresql-8.0.0.pctfree/src/include/storage/freespace.h    2005-08-20 23:18:46.661894056 +0900
***************
*** 39,49 ****
  extern void InitFreeSpaceMap(void);
  extern int    FreeSpaceShmemSize(void);

! extern BlockNumber GetPageWithFreeSpace(RelFileNode *rel, Size spaceNeeded);
  extern BlockNumber RecordAndGetPageWithFreeSpace(RelFileNode *rel,
                                BlockNumber oldPage,
                                Size oldSpaceAvail,
!                               Size spaceNeeded);
  extern Size GetAvgFSMRequestSize(RelFileNode *rel);
  extern void RecordRelationFreeSpace(RelFileNode *rel,
                          int nPages,
--- 39,52 ----
  extern void InitFreeSpaceMap(void);
  extern int    FreeSpaceShmemSize(void);

! extern BlockNumber GetPageWithFreeSpace(RelFileNode *rel,
!                                         Size spaceNeeded,
!                                         bool forInsert);
  extern BlockNumber RecordAndGetPageWithFreeSpace(RelFileNode *rel,
                                BlockNumber oldPage,
                                Size oldSpaceAvail,
!                               Size spaceNeeded,
!                                                  bool forInsert);
  extern Size GetAvgFSMRequestSize(RelFileNode *rel);
  extern void RecordRelationFreeSpace(RelFileNode *rel,
                          int nPages,