[Xen-devel] [PATCH RESEND 0/4] qemu-qdisk: Replace grant map by grant copy.

[Paulina Szubarczyk <paulinaszubarczyk@xxxxxxxxx>]

Hi,

The patches are resend with split of patch #2.

It is a proposition for implementation of the replacement of the grant map 
operation with grant copy. 

I would appreciate an opinion about the approach if is proper or maybe 
I assumed something wrongly, and if you see any possibility of improvement
or the things that need to be changed.
If the approach is any good, I need to still rethink batch mode, 
notification(?) 
and implementation for mini-os.

In the libs, gnttab, linbxc there is added interface and invocation of 
an ioctl(gntdev, IOCTL_GNTDEV_GRANT_COPY, ..) system call on the gnttdev 
device. 
Described in details in the following messages. It is not implemented for 
mini-os.

The grant map operation is replaced on the behalf of grant copy in 
qemu-xen-dir/hw/block/xen_disk. The implementation is described in the patch.

For the functional test I attached the device with a qdisk backend to the guest.
I successfully mounted it and stored files there. During creation of 
a file system on the device BLKIF_OP_DISCARD operation seems to fail(ret value 
different then zero) but it also fails for the original version due to error 
return from qemu.

I made fio tests before[0] and after[1] the changes with different iodepth and 
size of the block. The test which I run can be accessed on my github[2] but 
mainly after the warm up I run for 60 seconds:
    fio --time_based \
                --clocksource=clock_gettime \
                --rw=randread \
                --random_distribution=pareto:0.9 \
                --size=10g \
            --direct='1' \
            --ioengine=libaio \
                --filename=$DEV \
                --iodepth=$IODEPTH \
                --bs=$BS \
                --name=$NAME \
                --runtime=$RUNTIME >> $FILENAME
The test were repeated at least three times. 

Although before the changes results looks coherent for me, after there are
considerable peaks for iodepth = {4,8}.

[0] 
https://docs.google.com/spreadsheets/d/1n0lraodhF5jlNO-YWNTgl57Aoe3K5S7Qke8YQQkGCDQ/edit?usp=sharing
[1] 
https://docs.google.com/spreadsheets/d/1E6AMiB8ceJpExL6jWpH9u2yy6DZxzhmDUyFf-eUuJ0c/edit?usp=sharing
    - domU sheets
[2] https://github.com/paulina-szubarczyk/xen-benchmark
    - multitest_with_iodepth.sh


Thanks and regards, 
Paulina

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel