The patch titled
readahead: state based method
has been removed from the -mm tree. Its filename was
readahead-state-based-method.patch
This patch was dropped because an updated version will be merged
------------------------------------------------------
Subject: readahead: state based method
From: Wu Fengguang <wfg@xxxxxxxxxxxxxxxx>
This is the fast code path of adaptive read-ahead.
MAJOR STEPS
===========
- estimate a thrashing safe ra_size;
- assemble the next read-ahead request in file_ra_state;
- submit it.
THE REFERENCE MODEL
===================
1. inactive list has constant length and page flow speed
2. the observed stream receives a steady flow of read requests
3. no page activation, so that the inactive list forms a pipe
With that we get the picture showed below.
|<------------------------- constant length ------------------------->|
<<<<<<<<<<<<<<<<<<<<<<<<< steady flow of pages <<<<<<<<<<<<<<<<<<<<<<<<
+---------------------------------------------------------------------+
|tail inactive list head|
| ======= ==========---- |
| chunk A(stale pages) chunk B(stale + fresh pages) |
+---------------------------------------------------------------------+
REAL WORLD ISSUES
=================
Real world workloads will always have fluctuations (violation of assumption
1 and 2). To counteract it, a tunable parameter readahead_ratio is introduced
to make the estimation conservative enough. Violation of assumption 3 will
not lead to thrashing, it is there just for simplicity of discussion.
Signed-off-by: Wu Fengguang <wfg@xxxxxxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
mm/readahead.c | 169 +++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 169 insertions(+)
diff -puN mm/readahead.c~readahead-state-based-method mm/readahead.c
--- a/mm/readahead.c~readahead-state-based-method
+++ a/mm/readahead.c
@@ -23,6 +23,8 @@ void default_unplug_io_fn(struct backing
}
EXPORT_SYMBOL(default_unplug_io_fn);
+#include <asm/div64.h>
+
/*
* Convienent macros for min/max read-ahead pages.
* Note that MAX_RA_PAGES is rounded down, while MIN_RA_PAGES is rounded up.
@@ -929,6 +931,173 @@ static unsigned long ra_submit(struct fi
return actual;
}
+/*
+ * Deduce the read-ahead/look-ahead size from primitive values.
+ *
+ * Input:
+ * - @ra_size stores the estimated thrashing-threshold.
+ * - @la_size stores the look-ahead size of previous request.
+ */
+static int adjust_rala(unsigned long ra_max,
+ unsigned long *ra_size, unsigned long *la_size)
+{
+ /*
+ * Cancel asynchrous read-ahead,
+ * if there is a major upsurge of load, or fall of this stream's speed.
+ */
+ if (*ra_size <= *la_size * 2) {
+ ra_account(NULL, RA_EVENT_READAHEAD_SHRINK, *ra_size);
+ return 0;
+ }
+
+ /*
+ * Substract the old look-ahead to get real safe size for the next
+ * read-ahead request.
+ */
+ *ra_size -= *la_size;
+
+ /*
+ * Set new la_size according to the (still large) ra_size.
+ */
+ *la_size = *ra_size / LOOKAHEAD_RATIO;
+
+ return 1;
+}
+
+static void limit_rala(unsigned long ra_max, unsigned long la_old,
+ unsigned long *ra_size, unsigned long *la_size)
+{
+ unsigned long stream_shift;
+
+ /*
+ * Protect against too small I/O sizes,
+ * by mapping [0, 4*min] to [min, 4*min].
+ */
+ if (*ra_size < 4 * MIN_RA_PAGES)
+ *ra_size = MIN_RA_PAGES + *ra_size * 3 / 4;
+
+ /*
+ * Apply basic upper limits.
+ */
+ if (*ra_size > ra_max)
+ *ra_size = ra_max;
+ if (*la_size > *ra_size)
+ *la_size = *ra_size;
+
+ /*
+ * Make sure stream_shift is not too small.
+ * (So that the next global_shift will not be too small.)
+ */
+ stream_shift = la_old + (*ra_size - *la_size);
+ if (stream_shift < *ra_size / 4)
+ *la_size -= (*ra_size / 4 - stream_shift);
+}
+
+/*
+ * The function estimates two values:
+ * 1. thrashing-threshold for the current stream
+ * It is returned to make the next read-ahead request.
+ * 2. the remained safe space for the current chunk
+ * It will be checked to ensure that the current chunk is safe.
+ *
+ * The computation will be pretty accurate under heavy load, and will vibrate
+ * more on light load(with small global_shift), so the grow speed of ra_size
+ * must be limited, and a moderate large stream_shift must be insured.
+ *
+ * The following figure illustrates the formula used in the function:
+ * While the stream reads stream_shift pages inside the chunks,
+ * the chunks are shifted global_shift pages inside inactive_list.
+ * So
+ * thrashing_threshold = free_mem * stream_shift / global_shift;
+ *
+ *
+ * chunk A chunk B
+ * |<=============== global_shift ================|
+ * +-------------+ +-------------------+ |
+ * | # | | # | inactive_list |
+ * +-------------+ +-------------------+ head |
+ * |---->| |---------->|
+ * | |
+ * +-- stream_shift --+
+ */
+static unsigned long compute_thrashing_threshold(struct file_ra_state *ra,
+ unsigned long *remain)
+{
+ unsigned long global_size;
+ unsigned long global_shift;
+ unsigned long stream_shift;
+ unsigned long ra_size;
+ uint64_t ll;
+
+ global_size = nr_free_inactive_pages_node(numa_node_id());
+ global_shift = nr_scanned_pages_node(numa_node_id()) - ra->age;
+ global_shift |= 1UL;
+ stream_shift = ra_invoke_interval(ra);
+
+ /* future safe space */
+ ll = (uint64_t) stream_shift * global_size;
+ do_div(ll, global_shift);
+ ra_size = ll;
+
+ /* remained safe space */
+ if (global_size > global_shift) {
+ ll = (uint64_t) stream_shift * (global_size - global_shift);
+ do_div(ll, global_shift);
+ *remain = ll;
+ } else
+ *remain = 0;
+
+ ddprintk("compute_thrashing_threshold: "
+ "at %lu ra %lu=%lu*%lu/%lu, remain %lu for %lu\n",
+ ra->readahead_index, ra_size,
+ stream_shift, global_size, global_shift,
+ *remain, ra_lookahead_size(ra));
+
+ return ra_size;
+}
+
+/*
+ * Main function for file_ra_state based read-ahead.
+ */
+static unsigned long
+state_based_readahead(struct address_space *mapping, struct file *filp,
+ struct file_ra_state *ra,
+ struct page *page, pgoff_t offset,
+ unsigned long req_size, unsigned long ra_max)
+{
+ unsigned long ra_old, ra_size;
+ unsigned long la_old, la_size;
+ unsigned long remain_space;
+ unsigned long growth_limit;
+
+ la_old = la_size = ra->readahead_index - offset;
+ ra_old = ra_readahead_size(ra);
+ ra_size = compute_thrashing_threshold(ra, &remain_space);
+ ra_size = ra_size * readahead_ratio / 100;
+
+ if (page && remain_space <= la_size) {
+ rescue_pages(page, la_size);
+ return 0;
+ }
+
+ growth_limit = req_size;
+ growth_limit += ra_max / 16;
+ growth_limit += (2 + readahead_ratio / 64) * ra_old;
+ if (growth_limit > ra_max)
+ growth_limit = ra_max;
+
+ if (!adjust_rala(growth_limit, &ra_size, &la_size))
+ return 0;
+
+ limit_rala(growth_limit, la_old, &ra_size, &la_size);
+
+ ra_set_class(ra, RA_CLASS_STATE);
+ ra_set_index(ra, offset, ra->readahead_index);
+ ra_set_size(ra, ra_size, la_size);
+
+ return ra_submit(ra, mapping, filp);
+}
+
#endif /* CONFIG_ADAPTIVE_READAHEAD */
/*
_
Patches currently in -mm which might be from wfg@xxxxxxxxxxxxxxxx are
origin.patch
readahead-state-based-method.patch
readahead-state-based-method-check-node-id.patch
readahead-state-based-method-decouple-readahead_ratio-from-growth_limit.patch
readahead-state-based-method-cancel-lookahead-gracefully.patch
readahead-context-based-method.patch
readahead-initial-method-guiding-sizes.patch
readahead-initial-method-thrashing-guard-size.patch
readahead-initial-method-user-recommended-size.patch
readahead-initial-method.patch
readahead-backward-prefetching-method.patch
readahead-thrashing-recovery-method.patch
readahead-thrashing-recovery-method-check-unbalanced-aging.patch
readahead-thrashing-recovery-method-refill-holes.patch
readahead-call-scheme.patch
readahead-call-scheme-cleanup.patch
readahead-call-scheme-catch-thrashing-on-lookahead-time.patch
readahead-call-scheme-doc-fixes-for-readahead.patch
readahead-laptop-mode.patch
readahead-loop-case.patch
readahead-nfsd-case.patch
readahead-remove-parameter-ra_max-from-thrashing_recovery_readahead.patch
readahead-remove-parameter-ra_max-from-adjust_rala.patch
readahead-state-based-method-protect-against-tiny-size.patch
readahead-rename-state_based_readahead-to-clock_based_readahead.patch
readahead-account-i-o-block-times-for-stock-readahead.patch
readahead-rescue_pages-updates.patch
readahead-remove-noaction-shrink-events.patch
readahead-remove-size-limit-on-read_ahead_kb.patch
readahead-remove-size-limit-of-max_sectors_kb-on-read_ahead_kb.patch
readahead-partial-sendfile-fix.patch
readahead-turn-on-by-default.patch
-
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
