NDB tux optim 15 - fix wasted space in index node entries

ndb/src/kernel/blocks/dbtux/Dbtux.hpp

@@ -172,12 +172,21 @@ private:
    * Physical tuple address in TUP.  Provides fast access to table tuple
    * or index node.  Valid within the db node and across timeslices.
    * Not valid between db nodes or across restarts.
+   *
+   * To avoid wasting an Uint16 the pageid is split in two.
    */
   struct TupLoc {
-    Uint32 m_pageId;            // page i-value
+  private:
+    Uint16 m_pageId1;           // page i-value (big-endian)
+    Uint16 m_pageId2;
     Uint16 m_pageOffset;        // page offset in words
+  public:
     TupLoc();
     TupLoc(Uint32 pageId, Uint16 pageOffset);
+    Uint32 getPageId() const;
+    void setPageId(Uint32 pageId);
+    Uint32 getPageOffset() const;
+    void setPageOffset(Uint32 pageOffset);
     bool operator==(const TupLoc& loc) const;
     bool operator!=(const TupLoc& loc) const;
   };
@@ -224,18 +233,13 @@ private:
    * work entry                 part 5
    *
    * There are 3 links to other nodes: left child, right child, parent.
-   * These are in TupLoc format but the pageIds and pageOffsets are
-   * stored in separate arrays (saves 1 word).
-   *
    * Occupancy (number of entries) is at least 1 except temporarily when
-   * a node is about to be removed.  If occupancy is 1, only max entry
-   * is present but both min and max prefixes are set.
+   * a node is about to be removed.
    */
   struct TreeNode;
   friend struct TreeNode;
   struct TreeNode {
-    Uint32 m_linkPI[3];         // link to 0-left child 1-right child 2-parent
-    Uint16 m_linkPO[3];         // page offsets for above real page ids
+    TupLoc m_link[3];           // link to 0-left child 1-right child 2-parent
     unsigned m_side : 2;        // we are 0-left child 1-right child 2-root
     int m_balance : 2;          // balance -1, 0, +1
     unsigned pad1 : 4;
@@ -805,22 +809,52 @@ Dbtux::ConstData::operator=(Data data)
 
 inline
 Dbtux::TupLoc::TupLoc() :
-  m_pageId(RNIL),
+  m_pageId1(RNIL >> 16),
+  m_pageId2(RNIL & 0xFFFF),
   m_pageOffset(0)
 {
 }
 
 inline
 Dbtux::TupLoc::TupLoc(Uint32 pageId, Uint16 pageOffset) :
-  m_pageId(pageId),
+  m_pageId1(pageId >> 16),
+  m_pageId2(pageId & 0xFFFF),
   m_pageOffset(pageOffset)
 {
 }
 
+inline Uint32
+Dbtux::TupLoc::getPageId() const
+{
+  return (m_pageId1 << 16) | m_pageId2;
+}
+
+inline void
+Dbtux::TupLoc::setPageId(Uint32 pageId)
+{
+  m_pageId1 = (pageId >> 16);
+  m_pageId2 = (pageId & 0xFFFF);
+}
+
+inline Uint32
+Dbtux::TupLoc::getPageOffset() const
+{
+  return (Uint32)m_pageOffset;
+}
+
+inline void
+Dbtux::TupLoc::setPageOffset(Uint32 pageOffset)
+{
+  m_pageOffset = (Uint16)pageOffset;
+}
+
 inline bool
 Dbtux::TupLoc::operator==(const TupLoc& loc) const
 {
-  return m_pageId == loc.m_pageId && m_pageOffset == loc.m_pageOffset;
+  return
+    m_pageId1 == loc.m_pageId1 &&
+    m_pageId2 == loc.m_pageId2 &&
+    m_pageOffset == loc.m_pageOffset;
 }
 
 inline bool
@@ -851,13 +885,13 @@ Dbtux::TreeEnt::eq(const TreeEnt ent) const
 inline int
 Dbtux::TreeEnt::cmp(const TreeEnt ent) const
 {
-  if (m_tupLoc.m_pageId < ent.m_tupLoc.m_pageId)
+  if (m_tupLoc.getPageId() < ent.m_tupLoc.getPageId())
     return -1;
-  if (m_tupLoc.m_pageId > ent.m_tupLoc.m_pageId)
+  if (m_tupLoc.getPageId() > ent.m_tupLoc.getPageId())
     return +1;
-  if (m_tupLoc.m_pageOffset < ent.m_tupLoc.m_pageOffset)
+  if (m_tupLoc.getPageOffset() < ent.m_tupLoc.getPageOffset())
     return -1;
-  if (m_tupLoc.m_pageOffset > ent.m_tupLoc.m_pageOffset)
+  if (m_tupLoc.getPageOffset() > ent.m_tupLoc.getPageOffset())
     return +1;
   if (m_tupVersion < ent.m_tupVersion)
     return -1;
@@ -880,12 +914,9 @@ Dbtux::TreeNode::TreeNode() :
   m_occup(0),
   m_nodeScan(RNIL)
 {
-  m_linkPI[0] = NullTupLoc.m_pageId;
-  m_linkPO[0] = NullTupLoc.m_pageOffset;
-  m_linkPI[1] = NullTupLoc.m_pageId;
-  m_linkPO[1] = NullTupLoc.m_pageOffset;
-  m_linkPI[2] = NullTupLoc.m_pageId;
-  m_linkPO[2] = NullTupLoc.m_pageOffset;
+  m_link[0] = NullTupLoc;
+  m_link[1] = NullTupLoc;
+  m_link[2] = NullTupLoc;
 }
 
 // Dbtux::TreeHead
@@ -913,7 +944,6 @@ Dbtux::TreeHead::getSize(AccSize acc) const
   case AccFull:
     return m_nodeSize;
   }
-  abort();
   return 0;
 }
 
@@ -1088,13 +1118,13 @@ inline Dbtux::TupLoc
 Dbtux::NodeHandle::getLink(unsigned i)
 {
   ndbrequire(i <= 2);
-  return TupLoc(m_node->m_linkPI[i], m_node->m_linkPO[i]);
+  return m_node->m_link[i];
 }
 
 inline unsigned
 Dbtux::NodeHandle::getChilds()
 {
-  return (getLink(0) != NullTupLoc) + (getLink(1) != NullTupLoc);
+  return (m_node->m_link[0] != NullTupLoc) + (m_node->m_link[1] != NullTupLoc);
 }
 
 inline unsigned
@@ -1125,8 +1155,7 @@ inline void
 Dbtux::NodeHandle::setLink(unsigned i, TupLoc loc)
 {
   ndbrequire(i <= 2);
-  m_node->m_linkPI[i] = loc.m_pageId;
-  m_node->m_linkPO[i] = loc.m_pageOffset;
+  m_node->m_link[i] = loc;
 }
 
 inline void
@@ -1224,7 +1253,7 @@ Dbtux::getTupAddr(const Frag& frag, TreeEnt ent)
   const Uint32 tableFragPtrI = frag.m_tupTableFragPtrI[ent.m_fragBit];
   const TupLoc tupLoc = ent.m_tupLoc;
   Uint32 tupAddr = NullTupAddr;
-  c_tup->tuxGetTupAddr(tableFragPtrI, tupLoc.m_pageId, tupLoc.m_pageOffset, tupAddr);
+  c_tup->tuxGetTupAddr(tableFragPtrI, tupLoc.getPageId(), tupLoc.getPageOffset(), tupAddr);
   jamEntry();
   return tupAddr;
 }

ndb/src/kernel/blocks/dbtux/DbtuxDebug.cpp

@@ -256,8 +256,8 @@ operator<<(NdbOut& out, const Dbtux::TupLoc& loc)
   if (loc == Dbtux::NullTupLoc) {
     out << "null";
   } else {
-    out << dec << loc.m_pageId;
-    out << "." << dec << loc.m_pageOffset;
+    out << dec << loc.getPageId();
+    out << "." << dec << loc.getPageOffset();
   }
   return out;
 }
@@ -274,13 +274,10 @@ operator<<(NdbOut& out, const Dbtux::TreeEnt& ent)
 NdbOut&
 operator<<(NdbOut& out, const Dbtux::TreeNode& node)
 {
-  Dbtux::TupLoc link0(node.m_linkPI[0], node.m_linkPO[0]);
-  Dbtux::TupLoc link1(node.m_linkPI[1], node.m_linkPO[1]);
-  Dbtux::TupLoc link2(node.m_linkPI[2], node.m_linkPO[2]);
   out << "[TreeNode " << hex << &node;
-  out << " [left " << link0 << "]";
-  out << " [right " << link1 << "]";
-  out << " [up " << link2 << "]";
+  out << " [left " << node.m_link[0] << "]";
+  out << " [right " << node.m_link[1] << "]";
+  out << " [up " << node.m_link[2] << "]";
   out << " [side " << dec << node.m_side << "]";
   out << " [occup " << dec << node.m_occup << "]";
   out << " [balance " << dec << (int)node.m_balance << "]";

ndb/src/kernel/blocks/dbtux/DbtuxGen.cpp

@@ -245,7 +245,7 @@ Dbtux::readKeyAttrs(const Frag& frag, TreeEnt ent, unsigned start, Data keyData)
   const Uint32 numAttrs = frag.m_numAttrs - start;
   // skip to start position in keyAttrs only
   keyAttrs += start;
-  int ret = c_tup->tuxReadAttrs(tableFragPtrI, tupLoc.m_pageId, tupLoc.m_pageOffset, tupVersion, keyAttrs, numAttrs, keyData);
+  int ret = c_tup->tuxReadAttrs(tableFragPtrI, tupLoc.getPageId(), tupLoc.getPageOffset(), tupVersion, keyAttrs, numAttrs, keyData);
   jamEntry();
   // TODO handle error
   ndbrequire(ret > 0);
@@ -256,7 +256,7 @@ Dbtux::readTablePk(const Frag& frag, TreeEnt ent, Data pkData, unsigned& pkSize)
 {
   const Uint32 tableFragPtrI = frag.m_tupTableFragPtrI[ent.m_fragBit];
   const TupLoc tupLoc = ent.m_tupLoc;
-  int ret = c_tup->tuxReadPk(tableFragPtrI, tupLoc.m_pageId, tupLoc.m_pageOffset, pkData);
+  int ret = c_tup->tuxReadPk(tableFragPtrI, tupLoc.getPageId(), tupLoc.getPageOffset(), pkData);
   jamEntry();
   // TODO handle error
   ndbrequire(ret > 0);

ndb/src/kernel/blocks/dbtux/DbtuxMeta.cpp

@@ -235,6 +235,20 @@ Dbtux::execTUX_ADD_ATTRREQ(Signal* signal)
       tree.m_minOccup = tree.m_maxOccup - maxSlack;
       // root node does not exist (also set by ctor)
       tree.m_root = NullTupLoc;
+#ifdef VM_TRACE
+      if (debugFlags & DebugMeta) {
+        if (fragOpPtr.p->m_fragNo == 0) {
+          debugOut << "Index id=" << indexPtr.i;
+          debugOut << " nodeSize=" << tree.m_nodeSize;
+          debugOut << " headSize=" << NodeHeadSize;
+          debugOut << " prefSize=" << tree.m_prefSize;
+          debugOut << " entrySize=" << TreeEntSize;
+          debugOut << " minOccup=" << tree.m_minOccup;
+          debugOut << " maxOccup=" << tree.m_maxOccup;
+          debugOut << endl;
+        }
+      }
+#endif
       // fragment is defined
       c_fragOpPool.release(fragOpPtr);
     }

ndb/src/kernel/blocks/dbtux/DbtuxNode.cpp

@@ -24,8 +24,8 @@ int
 Dbtux::allocNode(Signal* signal, NodeHandle& node)
 {
   Frag& frag = node.m_frag;
-  Uint32 pageId = NullTupLoc.m_pageId;
-  Uint32 pageOffset = NullTupLoc.m_pageOffset;
+  Uint32 pageId = NullTupLoc.getPageId();
+  Uint32 pageOffset = NullTupLoc.getPageOffset();
   Uint32* node32 = 0;
   int errorCode = c_tup->tuxAllocNode(signal, frag.m_tupIndexFragPtrI, pageId, pageOffset, node32);
   jamEntry();
@@ -60,8 +60,8 @@ Dbtux::selectNode(Signal* signal, NodeHandle& node, TupLoc loc, AccSize acc)
 {
   Frag& frag = node.m_frag;
   ndbrequire(loc != NullTupLoc);
-  Uint32 pageId = loc.m_pageId;
-  Uint32 pageOffset = loc.m_pageOffset;
+  Uint32 pageId = loc.getPageId();
+  Uint32 pageOffset = loc.getPageOffset();
   Uint32* node32 = 0;
   c_tup->tuxGetNode(frag.m_tupIndexFragPtrI, pageId, pageOffset, node32);
   jamEntry();
@@ -100,8 +100,8 @@ Dbtux::deleteNode(Signal* signal, NodeHandle& node)
   Frag& frag = node.m_frag;
   ndbrequire(node.getOccup() == 0);
   TupLoc loc = node.m_loc;
-  Uint32 pageId = loc.m_pageId;
-  Uint32 pageOffset = loc.m_pageOffset;
+  Uint32 pageId = loc.getPageId();
+  Uint32 pageOffset = loc.getPageOffset();
   Uint32* node32 = reinterpret_cast<Uint32*>(node.m_node);
   c_tup->tuxFreeNode(signal, frag.m_tupIndexFragPtrI, pageId, pageOffset, node32);
   jamEntry();

ndb/src/kernel/blocks/dbtux/Times.txt

@@ -108,4 +108,11 @@ charsets        mc02/a  35 ms   60 ms    71 pct
 
 [ case b: TUX can no longer use pointers to TUP data ]
 
+optim 15        mc02/a  34 ms   60 ms    72 pct
+                mc02/b  42 ms   85 ms   100 pct
+                mc02/c   5 ms   12 ms   110 pct
+                mc02/d 178 ms  242 ms    35 pct
+
+[ corrected wasted space in index node ]
+
 vim: set et:

ndb/test/ndbapi/testOIBasic.cpp

@@ -212,6 +212,8 @@ struct Par : public Opt {
   // value calculation
   unsigned m_range;
   unsigned m_pctrange;
+  // choice of key
+  bool m_randomkey;
   // do verify after read
   bool m_verify;
   // deadlock possible
@@ -227,6 +229,7 @@ struct Par : public Opt {
     m_totrows(m_threads * m_rows),
     m_range(m_rows),
     m_pctrange(0),
+    m_randomkey(false),
     m_verify(false),
     m_deadlock(false) {
   }
@@ -2119,7 +2122,8 @@ pkupdate(Par par)
   Lst lst;
   bool deadlock = false;
   for (unsigned j = 0; j < par.m_rows; j++) {
-    unsigned i = thrrow(par, j);
+    unsigned j2 = ! par.m_randomkey ? j : urandom(par.m_rows);
+    unsigned i = thrrow(par, j2);
     set.lock();
     if (! set.exist(i) || set.pending(i)) {
       set.unlock();
@@ -2722,6 +2726,7 @@ pkupdateindexbuild(Par par)
   if (par.m_no == 0) {
     CHK(createindex(par) == 0);
   } else {
+    par.m_randomkey = true;
     CHK(pkupdate(par) == 0);
   }
   return 0;