runtime: finish sweeping before concurrent GC starts

Currently, the concurrent sweep follows a 1:1 rule: when allocation
needs a span, it sweeps a span (likewise, when a large allocation
needs N pages, it sweeps until it frees N pages). This rule worked
well for the STW collector (especially when GOGC==100) because it did
no more sweeping than necessary to keep the heap from growing, would
generally finish sweeping just before GC, and ensured good temporal
locality between sweeping a page and allocating from it.

It doesn't work well with concurrent GC. Since concurrent GC requires
starting GC earlier (sometimes much earlier), the sweep often won't be
done when GC starts. Unfortunately, the first thing GC has to do is
finish the sweep. In the mean time, the mutator can continue
allocating, pushing the heap size even closer to the goal size. This
worked okay with the 7/8ths trigger, but it gets into a vicious cycle
with the GC trigger controller: if the mutator is allocating quickly
and driving the trigger lower, more and more sweep work will be left
to GC; this both causes GC to take longer (allowing the mutator to
allocate more during GC) and delays the start of the concurrent mark
phase, which throws off the GC controller's statistics and generally
causes it to push the trigger even lower.

As an example of a particularly bad case, the garbage benchmark with
GOMAXPROCS=4 and -benchmem 512 (MB) spends the first 0.4-0.8 seconds
of each GC cycle sweeping, during which the heap grows by between
109MB and 252MB.

To fix this, this change replaces the 1:1 sweep rule with a
proportional sweep rule. At the end of GC, GC knows exactly how much
heap allocation will occur before the next concurrent GC as well as
how many span pages must be swept. This change computes this "sweep
ratio" and when the mallocgc asks for a span, the mcentral sweeps
enough spans to bring the swept span count into ratio with the
allocated byte count.

On the benchmark from above, this entirely eliminates sweeping at the
beginning of GC, which reduces the time between startGC readying the
GC goroutine and GC stopping the world for sweep termination to ~100µs
during which the heap grows at most 134KB.

Change-Id: I35422d6bba0c2310d48bb1f8f30a72d29e98c1af
Reviewed-on: https://go-review.googlesource.com/8921
Reviewed-by: Rick Hudson <rlh@golang.org>

src/runtime/mcentral.go

@@ -29,6 +29,24 @@ func mCentral_Init(c *mcentral, sizeclass int32) {
 
 // Allocate a span to use in an MCache.
 func mCentral_CacheSpan(c *mcentral) *mspan {
+	// Perform proportional sweep work. We don't directly reuse
+	// the spans we're sweeping here for this allocation because
+	// these can hold any size class. We'll sweep one more span
+	// below and use that because it will have the right size
+	// class and be hot in our cache.
+	pagesOwed := int64(mheap_.sweepPagesPerByte * float64(memstats.heap_live-memstats.heap_marked))
+	if pagesOwed-int64(mheap_.pagesSwept) > 1 {
+		// Get the debt down to one page, which we're likely
+		// to take care of below (if we don't, that's fine;
+		// we'll pick up the slack later).
+		for pagesOwed-int64(atomicload64(&mheap_.pagesSwept)) > 1 {
+			if gosweepone() == ^uintptr(0) {
+				mheap_.sweepPagesPerByte = 0
+				break
+			}
+		}
+	}
+
 	lock(&c.lock)
 	sg := mheap_.sweepgen
 retry:

src/runtime/mgc.go

@@ -580,12 +580,11 @@ func gc(mode int) {
 
 	// Pick up the remaining unswept/not being swept spans concurrently
 	//
-	// TODO(austin): If the last GC cycle shrank the heap, our 1:1
+	// This shouldn't happen if we're being invoked in background
-	// sweeping rule will undershoot and we'll wind up doing
+	// mode since proportional sweep should have just finished
-	// sweeping here, which will allow the mutator to do more
+	// sweeping everything, but rounding errors, etc, may leave a
-	// allocation than we intended before we "really" start GC.
+	// few spans unswept. In forced mode, this is necessary since
-	// Compute an allocation sweep ratio so we're done sweeping by
+	// GC can be forced at any point in the sweeping cycle.
-	// the time we hit next_gc.
 	for gosweepone() != ^uintptr(0) {
 		sweep.nbgsweep++
 	}
@@ -1025,6 +1024,11 @@ func gcSweep(mode int) {
 
 	if !_ConcurrentSweep || mode == gcForceBlockMode {
 		// Special case synchronous sweep.
+		// Record that no proportional sweeping has to happen.
+		lock(&mheap_.lock)
+		mheap_.sweepPagesPerByte = 0
+		mheap_.pagesSwept = 0
+		unlock(&mheap_.lock)
 		// Sweep all spans eagerly.
 		for sweepone() != ^uintptr(0) {
 			sweep.npausesweep++
@@ -1035,6 +1039,27 @@ func gcSweep(mode int) {
 		return
 	}
 
+	// Account how much sweeping needs to be done before the next
+	// GC cycle and set up proportional sweep statistics.
+	var pagesToSweep uintptr
+	for _, s := range work.spans {
+		if s.state == mSpanInUse {
+			pagesToSweep += s.npages
+		}
+	}
+	heapDistance := int64(memstats.next_gc) - int64(memstats.heap_live)
+	// Add a little margin so rounding errors and concurrent
+	// sweep are less likely to leave pages unswept when GC starts.
+	heapDistance -= 1024 * 1024
+	if heapDistance < _PageSize {
+		// Avoid setting the sweep ratio extremely high
+		heapDistance = _PageSize
+	}
+	lock(&mheap_.lock)
+	mheap_.sweepPagesPerByte = float64(pagesToSweep) / float64(heapDistance)
+	mheap_.pagesSwept = 0
+	unlock(&mheap_.lock)
+
 	// Background sweep.
 	lock(&sweep.lock)
 	if sweep.parked {

src/runtime/mgcsweep.go

@@ -165,6 +165,8 @@ func mSpan_Sweep(s *mspan, preserve bool) bool {
 		traceGCSweepStart()
 	}
 
+	xadd64(&mheap_.pagesSwept, int64(s.npages))
+
 	cl := s.sizeclass
 	size := s.elemsize
 	res := false

src/runtime/mheap.go

@@ -59,6 +59,10 @@ type mheap struct {
 	specialprofilealloc   fixalloc // allocator for specialprofile*
 	speciallock           mutex    // lock for sepcial record allocators.
 
+	// Proportional sweep
+	pagesSwept        uint64  // pages swept this cycle; updated atomically
+	sweepPagesPerByte float64 // proportional sweep ratio; written with lock, read without
+
 	// Malloc stats.
 	largefree  uint64                  // bytes freed for large objects (>maxsmallsize)
 	nlargefree uint64                  // number of frees for large objects (>maxsmallsize)
@@ -362,6 +366,13 @@ func mHeap_Alloc_m(h *mheap, npage uintptr, sizeclass int32, large bool) *mspan
 	// To prevent excessive heap growth, before allocating n pages
 	// we need to sweep and reclaim at least n pages.
 	if h.sweepdone == 0 {
+		// TODO(austin): This tends to sweep a large number of
+		// spans in order to find a few completely free spans
+		// (for example, in the garbage benchmark, this sweeps
+		// ~30x the number of pages its trying to allocate).
+		// If GC kept a bit for whether there were any marks
+		// in a span, we could release these free spans
+		// at the end of GC and eliminate this entirely.
 		mHeap_Reclaim(h, npage)
 	}