diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
new file mode 100644
index 8f467be..1a05666
*** a/src/backend/storage/lmgr/proc.c
--- b/src/backend/storage/lmgr/proc.c
*************** InitProcGlobal(void)
*** 168,173 ****
--- 168,177 ----
  	bool		found;
  	uint32		TotalProcs = MaxBackends + NUM_AUXILIARY_PROCS + max_prepared_xacts;
  
+ 	/* PGXACT should be aligned to the power of 2 */
+ 	StaticAssertExpr((sizeof(PGXACT) & (sizeof(PGXACT) - 1)) == 0,
+ 					 "PGXACT is not properly aligned");
+ 
  	/* Create the ProcGlobal shared structure */
  	ProcGlobal = (PROC_HDR *)
  		ShmemInitStruct("Proc Header", sizeof(PROC_HDR), &found);
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
new file mode 100644
index 5f38fa6..a7e0ea3
*** a/src/include/storage/proc.h
--- b/src/include/storage/proc.h
*************** extern PGDLLIMPORT struct PGXACT *MyPgXa
*** 189,199 ****
  /*
   * Prior to PostgreSQL 9.2, the fields below were stored as part of the
   * PGPROC.  However, benchmarking revealed that packing these particular
!  * members into a separate array as tightly as possible sped up GetSnapshotData
!  * considerably on systems with many CPU cores, by reducing the number of
!  * cache lines needing to be fetched.  Thus, think very carefully before adding
!  * anything else here.
   */
  typedef struct PGXACT
  {
  	TransactionId xid;			/* id of top-level transaction currently being
--- 189,208 ----
  /*
   * Prior to PostgreSQL 9.2, the fields below were stored as part of the
   * PGPROC.  However, benchmarking revealed that packing these particular
!  * members into a separate array sped up GetSnapshotData considerably on
!  * systems with many CPU cores, by reducing the number of cache lines needing
!  * to be fetched.  Also, this fields are subject of intensive writes even on
!  * read-only workloads.  Thereby, it's desirable that writes of these fields
!  * invalidate as less cache lines as possible.  From that point of view, we
!  * should prevent PGXACT from being on the boundary of cache lines.  In order
!  * to achieve that we align PGXACT to the nearest power of 2.  That gives
!  * considerable speedup on systems with many CPU cores.
   */
+ 
+ /* Calculation of padding for PGXACT.  Update this after changing of PGXACT. */
+ #define PGXACTPadSize (16 - 2 * sizeof(TransactionId) \
+ 					   - 2 * sizeof(uint8) - 2 * sizeof(bool))
+ 
  typedef struct PGXACT
  {
  	TransactionId xid;			/* id of top-level transaction currently being
*************** typedef struct PGXACT
*** 211,216 ****
--- 220,227 ----
  								 * previously called InCommit */
  
  	uint8		nxids;
+ 
+ 	char		pad[PGXACTPadSize];
  } PGXACT;
  
  /*