[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [PULL 22/38] block/parallels: improve image reading perform
From: |
Stefan Hajnoczi |
Subject: |
[Qemu-devel] [PULL 22/38] block/parallels: improve image reading performance |
Date: |
Fri, 22 May 2015 10:01:54 +0100 |
From: "Denis V. Lunev" <address@hidden>
Try to perform IO for the biggest continuous block possible.
The performance for sequential read is increased from 220 Mb/sec to
360 Mb/sec for continous image on my SSD HDD.
Signed-off-by: Denis V. Lunev <address@hidden>
Reviewed-by: Roman Kagan <address@hidden>
Reviewed-by: Stefan Hajnoczi <address@hidden>
Signed-off-by: Roman Kagan <address@hidden>
Message-id: address@hidden
CC: Kevin Wolf <address@hidden>
Signed-off-by: Stefan Hajnoczi <address@hidden>
---
block/parallels.c | 36 +++++++++++++++++++++++++++++++-----
1 file changed, 31 insertions(+), 5 deletions(-)
diff --git a/block/parallels.c b/block/parallels.c
index 76e3a4e..5ff74e8 100644
--- a/block/parallels.c
+++ b/block/parallels.c
@@ -103,6 +103,35 @@ static int cluster_remainder(BDRVParallelsState *s,
int64_t sector_num,
return MIN(nb_sectors, ret);
}
+static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
+ int nb_sectors, int *pnum)
+{
+ int64_t start_off = -2, prev_end_off = -2;
+
+ *pnum = 0;
+ while (nb_sectors > 0 || start_off == -2) {
+ int64_t offset = seek_to_sector(s, sector_num);
+ int to_end;
+
+ if (start_off == -2) {
+ start_off = offset;
+ prev_end_off = offset;
+ } else if (offset != prev_end_off) {
+ break;
+ }
+
+ to_end = cluster_remainder(s, sector_num, nb_sectors);
+ nb_sectors -= to_end;
+ sector_num += to_end;
+ *pnum += to_end;
+
+ if (offset > 0) {
+ prev_end_off += to_end;
+ }
+ }
+ return start_off;
+}
+
static int64_t allocate_cluster(BlockDriverState *bs, int64_t sector_num)
{
BDRVParallelsState *s = bs->opaque;
@@ -148,11 +177,9 @@ static int64_t coroutine_fn
parallels_co_get_block_status(BlockDriverState *bs,
int64_t offset;
qemu_co_mutex_lock(&s->lock);
- offset = seek_to_sector(s, sector_num);
+ offset = block_status(s, sector_num, nb_sectors, pnum);
qemu_co_mutex_unlock(&s->lock);
- *pnum = cluster_remainder(s, sector_num, nb_sectors);
-
if (offset < 0) {
return 0;
}
@@ -218,10 +245,9 @@ static coroutine_fn int
parallels_co_readv(BlockDriverState *bs,
int n, nbytes;
qemu_co_mutex_lock(&s->lock);
- position = seek_to_sector(s, sector_num);
+ position = block_status(s, sector_num, nb_sectors, &n);
qemu_co_mutex_unlock(&s->lock);
- n = cluster_remainder(s, sector_num, nb_sectors);
nbytes = n << BDRV_SECTOR_BITS;
if (position < 0) {
--
2.1.0
- [Qemu-devel] [PULL 13/38] parallels: change copyright information in the image header, (continued)
- [Qemu-devel] [PULL 13/38] parallels: change copyright information in the image header, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 14/38] block/parallels: rename catalog_ names to bat_, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 18/38] block/parallels: move parallels_open/probe to the very end of the file, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 16/38] block/parallels: keep BAT bitmap data in little endian in memory, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 15/38] block/parallels: create bat2sect helper, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 19/38] block/parallels: implement parallels_check method of block driver, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 17/38] block/parallels: read parallels image header and BAT into single buffer, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 20/38] block/parallels: implement incorrect close detection, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 21/38] iotests, parallels: check for incorrectly closed image in tests, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 24/38] block/parallels: delay writing to BAT till bdrv_co_flush_to_os, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 22/38] block/parallels: improve image reading performance,
Stefan Hajnoczi <=
- [Qemu-devel] [PULL 23/38] block/parallels: create bat_entry_off helper, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 26/38] block/parallels: optimize linear image expansion, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 28/38] configure: handle clang -nopie argument warning, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 27/38] block/parallels: improve image writing performance further, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 29/38] configure: factor out supported flag check, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 25/38] block/parallels: add prealloc-mode and prealloc-size open paramemets, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 30/38] configure: silence glib unknown attribute __alloc_size__, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 31/38] configure: Add workaround for ccache and clang, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 33/38] block: minimal bounce buffer alignment, Stefan Hajnoczi, 2015/05/22
- [Qemu-devel] [PULL 34/38] block: align bounce buffers to page, Stefan Hajnoczi, 2015/05/22