More testing on AIO with O_DIRECT and /dev/raw/. Doing AIO reads was
using a lot more cpu time than using dd on the raw partition even with
the io_queue_wait patch. Found out that aio is doing readahead even
for O_DIRECT. Here's a patch that fixes it. Here are some output
from time on reading a 90mb partition on a raw device with 32k i/o:
CMD Kernel I/O real user sys
RUN version size
=== ====== === ==== ==== ===
dd 2.5.73-mm2: 32k 2.514s 0.000s 0.200s
aio (1 iocb) 2.5.73-mm2: 32k 3.408s 1.120s 2.250s
aio (32 iocb) 2.5.73-mm2: 32k 2.994s 0.830s 2.160s
aio (1 iocb) 2.5.73-mm2+patch.aiodio 32k 2.509s 0.850s 1.660s
aio (1 iocb) 2.5.73-mm2+patch.aiodio
+patch.io_queue_wait 32k 2.566s 0.010s 0.240s
aio (32 iocb) 2.5.73-mm2+patch.aiodio
+patch.io_queue_wait 32k 2.465s 0.010s 0.080s
With this patch and my previous io_queue_wait patch the cpu time
drops down to very little when doing AIO with O_DIRECT.
Daniel McNeil <daniel@osdl.org>
--- linux-2.5.73-mm2/fs/aio.c 2003-06-30 16:39:15.874216228 -0700
+++ linux-2.5.73-mm2.aiodio/fs/aio.c 2003-06-30 15:38:46.000000000 -0700
@@ -1380,15 +1380,20 @@ ssize_t aio_setup_iocb(struct kiocb *kio
break;
ret = -EINVAL;
if (file->f_op->aio_read) {
- struct address_space *mapping =
- file->f_dentry->d_inode->i_mapping;
- unsigned long index = kiocb->ki_pos >> PAGE_CACHE_SHIFT;
- unsigned long end = (kiocb->ki_pos + kiocb->ki_left)
- >> PAGE_CACHE_SHIFT;
-
- for (; index < end; index++) {
- page_cache_readahead(mapping, &file->f_ra,
- file, index);
+ /*
+ * Do not do readahead for DIRECT i/o
+ */
+ if (!(file->f_flags & O_DIRECT)) {
+ struct address_space *mapping =
+ file->f_dentry->d_inode->i_mapping;
+ unsigned long index = kiocb->ki_pos >> PAGE_CACHE_SHIFT;
+ unsigned long end = (kiocb->ki_pos + kiocb->ki_left)
+ >> PAGE_CACHE_SHIFT;
+
+ for (; index < end; index++) {
+ page_cache_readahead(mapping, &file->f_ra,
+ file, index);
+ }
}
kiocb->ki_retry = aio_pread;
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Jun 30 2003 - 22:00:35 EST