From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sat, 23 Nov 2024 21:47:10 +0000 (-0500)
Subject: bcachefs: Bias reads more in favor of faster device
X-Git-Tag: v6.14-rc1~204^2~156
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=e1cb4f56dc4cfcf1cb1b90479cb35c5a304ff527;p=thirdparty%2Fkernel%2Flinux.git

bcachefs: Bias reads more in favor of faster device

Per reports of performance issues on mixed multi device filesystems
where we're issuing too much IO to the spinning rust - tweak this
algorithm.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---

diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 98bb680b3860b..83aeceb688479 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -89,6 +89,14 @@ static inline bool ptr_better(struct bch_fs *c,
 		u64 l1 = dev_latency(c, p1.ptr.dev);
 		u64 l2 = dev_latency(c, p2.ptr.dev);
 
+		/*
+		 * Square the latencies, to bias more in favor of the faster
+		 * device - we never want to stop issuing reads to the slower
+		 * device altogether, so that we can update our latency numbers:
+		 */
+		l1 *= l1;
+		l2 *= l2;
+
 		/* Pick at random, biased in favor of the faster device: */
 
 		return bch2_rand_range(l1 + l2) > l1;