[Qemu-devel] [RFC][PATCH 1.8] block/iscsi: set bdi->cluster_size

[Qemu-devel] [RFC][PATCH 1.8] block/iscsi: set bdi->cluster_size

[Peter Lieven]

this patch aims to set bdi->cluster_size to the internal page size
of the iscsi target so that enabled callers can align requests
properly.

Signed-off-by: Peter Lieven <pl@kamp.de>
---
 block/iscsi.c |    6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/block/iscsi.c b/block/iscsi.c
index b7b5238..3a4c5d4 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -1569,6 +1569,12 @@ static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
     IscsiLun *iscsilun = bs->opaque;
     bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
     bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
+    bdi->cluster_size = iscsilun->bl.opt_xfer_len * iscsilun->block_size;
+    /* if opt_xfer_len is not specified guess the internal page size of the
+     * iscsi target by the means of opt_unmap_gran */
+    if (bdi->cluster_size == 0) {
+        bdi->cluster_size = iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
+    }
     return 0;
 }
 
-- 
1.7.9.5

Re: [Qemu-devel] [RFC][PATCH 1.8] block/iscsi: set bdi->cluster_size

[Paolo Bonzini]

Il 25/11/2013 09:29, Peter Lieven ha scritto:
> this patch aims to set bdi->cluster_size to the internal page size
> of the iscsi target so that enabled callers can align requests
> properly.
> 
> Signed-off-by: Peter Lieven <pl@kamp.de>
> ---
>  block/iscsi.c |    6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/block/iscsi.c b/block/iscsi.c
> index b7b5238..3a4c5d4 100644
> --- a/block/iscsi.c
> +++ b/block/iscsi.c
> @@ -1569,6 +1569,12 @@ static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
>      IscsiLun *iscsilun = bs->opaque;
>      bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
>      bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
> +    bdi->cluster_size = iscsilun->bl.opt_xfer_len * iscsilun->block_size;
> +    /* if opt_xfer_len is not specified guess the internal page size of the
> +     * iscsi target by the means of opt_unmap_gran */
> +    if (bdi->cluster_size == 0) {
> +        bdi->cluster_size = iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
> +    }
>      return 0;
>  }
>  
> 

This is not what cluster_size does; cluster_size is the granularity of
allocations, and hence COW requests.  In that sense, opt_unmap_gran is
probably a better choice than opt_xfer_len, especially if you want to
set the default cluster size of a new image (e.g. qcow2) so that the
format itself will not try to discard partial clusters.

However, I'm not sure overloading cluster_size is a good thing; you can
already use discard_alignment.  Also, I think you need some clamping
(e.g. if <4KB or >1MB).

Paolo

Re: [Qemu-devel] [RFC][PATCH 1.8] block/iscsi: set bdi->cluster_size

[Peter Lieven]

On 25.11.2013 09:54, Paolo Bonzini wrote:
> Il 25/11/2013 09:29, Peter Lieven ha scritto:
>> this patch aims to set bdi->cluster_size to the internal page size
>> of the iscsi target so that enabled callers can align requests
>> properly.
>>
>> Signed-off-by: Peter Lieven <pl@kamp.de>
>> ---
>>   block/iscsi.c |    6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/block/iscsi.c b/block/iscsi.c
>> index b7b5238..3a4c5d4 100644
>> --- a/block/iscsi.c
>> +++ b/block/iscsi.c
>> @@ -1569,6 +1569,12 @@ static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
>>       IscsiLun *iscsilun = bs->opaque;
>>       bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
>>       bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
>> +    bdi->cluster_size = iscsilun->bl.opt_xfer_len * iscsilun->block_size;
>> +    /* if opt_xfer_len is not specified guess the internal page size of the
>> +     * iscsi target by the means of opt_unmap_gran */
>> +    if (bdi->cluster_size == 0) {
>> +        bdi->cluster_size = iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
>> +    }
>>       return 0;
>>   }
>>   
>>
> This is not what cluster_size does; cluster_size is the granularity of
> allocations, and hence COW requests.  In that sense, opt_unmap_gran is
> probably a better choice than opt_xfer_len, especially if you want to
> set the default cluster size of a new image (e.g. qcow2) so that the
> format itself will not try to discard partial clusters.
>
> However, I'm not sure overloading cluster_size is a good thing; you can
> already use discard_alignment.  Also, I think you need some clamping
> (e.g. if <4KB or >1MB).
Maybe I should explain the background of this patch. I currently work
on some patches optimizing qemu-img convert. And I found out
that there is significant speed up if the writes to the output
image are aligned to the underlying page boundaries of the iscsi_target.
So I wanted to use bdi->cluster_size for aligning the
requests if its available. The same that already happens for compressed
images.

Maybe I should put together the whole series maybe this gives a better
picture.

Peter