diff --git a/NOTICE.txt b/NOTICE.txt
index fb16a280ccd..71efa526fbc 100644
--- a/NOTICE.txt
+++ b/NOTICE.txt
@@ -38,8 +38,10 @@ Copyright Jan Kovařík
 Licensed under the Apache License v2.0 as a part of the Bootstrap project.
 
 --
-This product includes portions of the Guava project v14, specifically
+This product includes portions of the Guava project v14 and v21, specifically
 'hbase-common/src/main/java/org/apache/hadoop/hbase/io/LimitInputStream.java'
+'hbase-common/src/main/java/org/apache/hadoop/hbase/util/Bytes.java'
+'hbase-common/src/main/java/org/apache/hadoop/hbase/util/ByteBufferUtils.java'
 
 Copyright (C) 2007 The Guava Authors
 
diff --git a/hbase-common/src/main/java/org/apache/hadoop/hbase/util/ByteBufferUtils.java b/hbase-common/src/main/java/org/apache/hadoop/hbase/util/ByteBufferUtils.java
index 34a4e02497b..6dac30004b0 100644
--- a/hbase-common/src/main/java/org/apache/hadoop/hbase/util/ByteBufferUtils.java
+++ b/hbase-common/src/main/java/org/apache/hadoop/hbase/util/ByteBufferUtils.java
@@ -701,46 +701,43 @@ public final class ByteBufferUtils {
   }
 
   static int compareToUnsafe(Object obj1, long o1, int l1, Object obj2, long o2, int l2) {
+    final int stride = 8;
     final int minLength = Math.min(l1, l2);
-    final int minWords = minLength / Bytes.SIZEOF_LONG;
+    int strideLimit = minLength & ~(stride - 1);
+    int i;
 
     /*
      * Compare 8 bytes at a time. Benchmarking shows comparing 8 bytes at a time is no slower than
      * comparing 4 bytes at a time even on 32-bit. On the other hand, it is substantially faster on
      * 64-bit.
      */
-    int j = minWords << 3; // Same as minWords * SIZEOF_LONG
-    for (int i = 0; i < j; i += Bytes.SIZEOF_LONG) {
-      long lw = UnsafeAccess.theUnsafe.getLong(obj1, o1 + i);
-      long rw = UnsafeAccess.theUnsafe.getLong(obj2, o2 + i);
-      long diff = lw ^ rw;
-      if (diff != 0) {
-        return lessThanUnsignedLong(lw, rw) ? -1 : 1;
-      }
-    }
-    int offset = j;
+    for (i = 0; i < strideLimit; i += stride) {
+      long lw = UnsafeAccess.theUnsafe.getLong(obj1, o1 + (long) i);
+      long rw = UnsafeAccess.theUnsafe.getLong(obj2, o2 + (long) i);
+      if (lw != rw) {
+        if (!UnsafeAccess.littleEndian) {
+          return ((lw + Long.MIN_VALUE) < (rw + Long.MIN_VALUE)) ? -1 : 1;
+        }
 
-    if (minLength - offset >= Bytes.SIZEOF_INT) {
-      int il = UnsafeAccess.theUnsafe.getInt(obj1, o1 + offset);
-      int ir = UnsafeAccess.theUnsafe.getInt(obj2, o2 + offset);
+        /*
+         * We want to compare only the first index where left[index] != right[index]. This
+         * corresponds to the least significant nonzero byte in lw ^ rw, since lw and rw are
+         * little-endian. Long.numberOfTrailingZeros(diff) tells us the least significant
+         * nonzero bit, and zeroing out the first three bits of L.nTZ gives us the shift to get
+         * that least significant nonzero byte. This comparison logic is based on UnsignedBytes
+         * from guava v21
+         */
+        int n = Long.numberOfTrailingZeros(lw ^ rw) & ~0x7;
+        return ((int) ((lw >>> n) & 0xFF)) - ((int) ((rw >>> n) & 0xFF));
+      }
+    }
+
+    // The epilogue to cover the last (minLength % stride) elements.
+    for (; i < minLength; i++) {
+      int il = (UnsafeAccess.theUnsafe.getByte(obj1, o1 + i) & 0xFF);
+      int ir = (UnsafeAccess.theUnsafe.getByte(obj2, o2 + i) & 0xFF);
       if (il != ir) {
-        return lessThanUnsignedInt(il, ir) ? -1 : 1;
-      }
-      offset += Bytes.SIZEOF_INT;
-    }
-    if (minLength - offset >= Bytes.SIZEOF_SHORT) {
-      short sl = UnsafeAccess.theUnsafe.getShort(obj1, o1 + offset);
-      short sr = UnsafeAccess.theUnsafe.getShort(obj2, o2 + offset);
-      if (sl != sr) {
-        return lessThanUnsignedShort(sl, sr) ? -1 : 1;
-      }
-      offset += Bytes.SIZEOF_SHORT;
-    }
-    if (minLength - offset == 1) {
-      int a = (UnsafeAccess.theUnsafe.getByte(obj1, o1 + offset) & 0xff);
-      int b = (UnsafeAccess.theUnsafe.getByte(obj2, o2 + offset) & 0xff);
-      if (a != b) {
-        return a - b;
+        return il - ir;
       }
     }
     return l1 - l2;
diff --git a/hbase-common/src/main/java/org/apache/hadoop/hbase/util/Bytes.java b/hbase-common/src/main/java/org/apache/hadoop/hbase/util/Bytes.java
index 704d97ffa38..a481d27a6b9 100644
--- a/hbase-common/src/main/java/org/apache/hadoop/hbase/util/Bytes.java
+++ b/hbase-common/src/main/java/org/apache/hadoop/hbase/util/Bytes.java
@@ -1575,47 +1575,42 @@ public class Bytes implements Comparable<Bytes> {
             length1 == length2) {
           return 0;
         }
+        final int stride = 8;
         final int minLength = Math.min(length1, length2);
-        final int minWords = minLength / SIZEOF_LONG;
+        int strideLimit = minLength & ~(stride - 1);
         final long offset1Adj = offset1 + UnsafeAccess.BYTE_ARRAY_BASE_OFFSET;
         final long offset2Adj = offset2 + UnsafeAccess.BYTE_ARRAY_BASE_OFFSET;
+        int i;
 
         /*
-         * Compare 8 bytes at a time. Benchmarking shows comparing 8 bytes at a
-         * time is no slower than comparing 4 bytes at a time even on 32-bit.
-         * On the other hand, it is substantially faster on 64-bit.
+         * Compare 8 bytes at a time. Benchmarking on x86 shows a stride of 8 bytes is no slower
+         * than 4 bytes even on 32-bit. On the other hand, it is substantially faster on 64-bit.
          */
-        // This is the end offset of long parts.
-        int j = minWords << 3; // Same as minWords * SIZEOF_LONG
-        for (int i = 0; i < j; i += SIZEOF_LONG) {
+        for (i = 0; i < strideLimit; i += stride) {
           long lw = theUnsafe.getLong(buffer1, offset1Adj + (long) i);
           long rw = theUnsafe.getLong(buffer2, offset2Adj + (long) i);
-          long diff = lw ^ rw;
-          if (diff != 0) {
-              return lessThanUnsignedLong(lw, rw) ? -1 : 1;
-          }
-        }
-        int offset = j;
+          if (lw != rw) {
+            if(!UnsafeAccess.littleEndian) {
+              return ((lw + Long.MIN_VALUE) < (rw + Long.MIN_VALUE)) ? -1 : 1;
+            }
 
-        if (minLength - offset >= SIZEOF_INT) {
-          int il = theUnsafe.getInt(buffer1, offset1Adj + offset);
-          int ir = theUnsafe.getInt(buffer2, offset2Adj + offset);
-          if (il != ir) {
-            return lessThanUnsignedInt(il, ir) ? -1: 1;
+            /*
+             * We want to compare only the first index where left[index] != right[index]. This
+             * corresponds to the least significant nonzero byte in lw ^ rw, since lw and rw are
+             * little-endian. Long.numberOfTrailingZeros(diff) tells us the least significant
+             * nonzero bit, and zeroing out the first three bits of L.nTZ gives us the shift to get
+             * that least significant nonzero byte. This comparison logic is based on UnsignedBytes
+             * comparator from guava v21
+             */
+            int n = Long.numberOfTrailingZeros(lw ^ rw) & ~0x7;
+            return ((int) ((lw >>> n) & 0xFF)) - ((int) ((rw >>> n) & 0xFF));
           }
-          offset += SIZEOF_INT;
         }
-        if (minLength - offset >= SIZEOF_SHORT) {
-          short sl = theUnsafe.getShort(buffer1, offset1Adj + offset);
-          short sr = theUnsafe.getShort(buffer2, offset2Adj + offset);
-          if (sl != sr) {
-            return lessThanUnsignedShort(sl, sr) ? -1: 1;
-          }
-          offset += SIZEOF_SHORT;
-        }
-        if (minLength - offset == 1) {
-          int a = (buffer1[(int)(offset1 + offset)] & 0xff);
-          int b = (buffer2[(int)(offset2 + offset)] & 0xff);
+
+        // The epilogue to cover the last (minLength % stride) elements.
+        for (; i < minLength; i++) {
+          int a = (buffer1[offset1 + i] & 0xFF);
+          int b = (buffer2[offset2 + i] & 0xFF);
           if (a != b) {
             return a - b;
           }