diff --git a/compiler/jit/jit_compiler.cc b/compiler/jit/jit_compiler.cc
index 23601c3..79a6d38 100644
--- a/compiler/jit/jit_compiler.cc
+++ b/compiler/jit/jit_compiler.cc
@@ -230,10 +230,10 @@
   }
 
   // Trim maps to reduce memory usage.
-  // TODO: measure how much this increases compile time.
+  // TODO: move this to an idle phase.
   {
     TimingLogger::ScopedTiming t2("TrimMaps", &logger);
-    runtime->GetArenaPool()->TrimMaps();
+    runtime->GetJitArenaPool()->TrimMaps();
   }
 
   total_time_ += NanoTime() - start_time;
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 5a9f258..13d6d62 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -861,7 +861,7 @@
   const uint32_t access_flags = method->GetAccessFlags();
   const InvokeType invoke_type = method->GetInvokeType();
 
-  ArenaAllocator arena(Runtime::Current()->GetArenaPool());
+  ArenaAllocator arena(Runtime::Current()->GetJitArenaPool());
   CodeVectorAllocator code_allocator(&arena);
   std::unique_ptr<CodeGenerator> codegen;
   {
diff --git a/runtime/base/arena_allocator.cc b/runtime/base/arena_allocator.cc
index a4b38ea..44af3f7 100644
--- a/runtime/base/arena_allocator.cc
+++ b/runtime/base/arena_allocator.cc
@@ -183,10 +183,10 @@
   free(reinterpret_cast<void*>(memory_));
 }
 
-MemMapArena::MemMapArena(size_t size, bool low_4gb) {
+MemMapArena::MemMapArena(size_t size, bool low_4gb, const char* name) {
   std::string error_msg;
   map_.reset(MemMap::MapAnonymous(
-      "LinearAlloc", nullptr, size, PROT_READ | PROT_WRITE, low_4gb, false, &error_msg));
+      name, nullptr, size, PROT_READ | PROT_WRITE, low_4gb, false, &error_msg));
   CHECK(map_.get() != nullptr) << error_msg;
   memory_ = map_->Begin();
   size_ = map_->Size();
@@ -210,9 +210,12 @@
   }
 }
 
-ArenaPool::ArenaPool(bool use_malloc, bool low_4gb)
-    : use_malloc_(use_malloc), lock_("Arena pool lock", kArenaPoolLock), free_arenas_(nullptr),
-      low_4gb_(low_4gb) {
+ArenaPool::ArenaPool(bool use_malloc, bool low_4gb, const char* name)
+    : use_malloc_(use_malloc),
+      lock_("Arena pool lock", kArenaPoolLock),
+      free_arenas_(nullptr),
+      low_4gb_(low_4gb),
+      name_(name) {
   if (low_4gb) {
     CHECK(!use_malloc) << "low4gb must use map implementation";
   }
@@ -250,7 +253,7 @@
   }
   if (ret == nullptr) {
     ret = use_malloc_ ? static_cast<Arena*>(new MallocArena(size)) :
-        new MemMapArena(size, low_4gb_);
+        new MemMapArena(size, low_4gb_, name_);
   }
   ret->Reset();
   return ret;
diff --git a/runtime/base/arena_allocator.h b/runtime/base/arena_allocator.h
index 8a96571..728f897 100644
--- a/runtime/base/arena_allocator.h
+++ b/runtime/base/arena_allocator.h
@@ -261,7 +261,7 @@
 
 class MemMapArena FINAL : public Arena {
  public:
-  MemMapArena(size_t size, bool low_4gb);
+  MemMapArena(size_t size, bool low_4gb, const char* name);
   virtual ~MemMapArena();
   void Release() OVERRIDE;
 
@@ -271,7 +271,9 @@
 
 class ArenaPool {
  public:
-  explicit ArenaPool(bool use_malloc = true, bool low_4gb = false);
+  ArenaPool(bool use_malloc = true,
+            bool low_4gb = false,
+            const char* name = "LinearAlloc");
   ~ArenaPool();
   Arena* AllocArena(size_t size) REQUIRES(!lock_);
   void FreeArenaChain(Arena* first) REQUIRES(!lock_);
@@ -287,6 +289,7 @@
   mutable Mutex lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
   Arena* free_arenas_ GUARDED_BY(lock_);
   const bool low_4gb_;
+  const char* name_;
   DISALLOW_COPY_AND_ASSIGN(ArenaPool);
 };
 
diff --git a/runtime/runtime.cc b/runtime/runtime.cc
index eb5455a..47ef214 100644
--- a/runtime/runtime.cc
+++ b/runtime/runtime.cc
@@ -316,6 +316,7 @@
   linear_alloc_.reset();
   low_4gb_arena_pool_.reset();
   arena_pool_.reset();
+  jit_arena_pool_.reset();
   MemMap::Shutdown();
   ATRACE_END();
 
@@ -1019,10 +1020,13 @@
   // Use MemMap arena pool for jit, malloc otherwise. Malloc arenas are faster to allocate but
   // can't be trimmed as easily.
   const bool use_malloc = IsAotCompiler();
-  arena_pool_.reset(new ArenaPool(use_malloc, false));
+  arena_pool_.reset(new ArenaPool(use_malloc, /* low_4gb */ false));
+  jit_arena_pool_.reset(
+      new ArenaPool(/* use_malloc */ false, /* low_4gb */ false, "CompilerMetadata"));
+
   if (IsAotCompiler() && Is64BitInstructionSet(kRuntimeISA)) {
     // 4gb, no malloc. Explanation in header.
-    low_4gb_arena_pool_.reset(new ArenaPool(false, true));
+    low_4gb_arena_pool_.reset(new ArenaPool(/* use_malloc */ false, /* low_4gb */ true));
   }
   linear_alloc_.reset(CreateLinearAlloc());
 
diff --git a/runtime/runtime.h b/runtime/runtime.h
index 8aac4ce..83e77d2 100644
--- a/runtime/runtime.h
+++ b/runtime/runtime.h
@@ -561,6 +561,9 @@
   ArenaPool* GetArenaPool() {
     return arena_pool_.get();
   }
+  ArenaPool* GetJitArenaPool() {
+    return jit_arena_pool_.get();
+  }
   const ArenaPool* GetArenaPool() const {
     return arena_pool_.get();
   }
@@ -669,6 +672,7 @@
 
   gc::Heap* heap_;
 
+  std::unique_ptr<ArenaPool> jit_arena_pool_;
   std::unique_ptr<ArenaPool> arena_pool_;
   // Special low 4gb pool for compiler linear alloc. We need ArtFields to be in low 4gb if we are
   // compiling using a 32 bit image on a 64 bit compiler in case we resolve things in the image