[Virtio-fs] [PATCH v3 00/13] virtio-fs: shared file system for virtual machines
vgoyal at redhat.com
Wed Aug 21 17:37:29 UTC 2019
Here are the V3 patches for virtio-fs filesystem. This time I have
broken the patch series in two parts. This is first part which does
not contain DAX support. Second patch series will contain the patches
for DAX support.
I have also dropped RFC tag from first patch series as we believe its
in good enough shape that it should get a consideration for inclusion
These patches apply on top of 5.3-rc5 kernel and are also available
Patches for V1 and V2 were posted here.
More information about the project can be found here.
Changes from V2
- Various bug fixes and performance improvements.
We have put instructions on how to use it here.
Some Performance Numbers
I have basically run bunch of fio jobs to get a sense of speed of
various operations. I wrote a simple wrapper script to run fio jobs
3 times and take their average and report it. These scripts are available
I set up a directory on ramfs on host and exported that directory inside
guest using virtio-9p and virtio-fs and ran tests inside guests. Ran
tests with cache=none both for virtio-9p and virtio-fs so that no caching
happens in guest. For virtio-fs, I ran an additional set of tests with
dax enabled. Dax is not part of first patch series but I included
results here because dax seems to get the maximum performance advantage
and its shows the real potential of virtio-fs.
- A fedora 28 host with 32G RAM, 2 sockets (6 cores per socket, 2
threads per core)
- Using ramfs on host as backing store. 4 fio files of 2G each.
- Created a VM with 16 VCPUS and 8GB memory. An 8GB cache window (for dax
- Results in three configurations have been reported. 9p (cache=none),
virtio-fs (cache=none) and virtio-fs (cache=none + dax).
There are other caching modes as well but to me cache=none seemed most
interesting for now because it does not cache anything in guest
and provides strong coherence. Other modes which provide less strong
coherence and hence are faster are yet to be benchmarked.
- Three fio ioengines psync, libaio and mmap have been used.
- I/O Workload of randread, radwrite, seqread and seqwrite have been run.
- Each file size is 2G. Block size 4K. iodepth=16
- "multi" means same operation was done with 4 jobs and each job is
operating on a file of size 2G.
- Some results are "0 (KiB/s)". That means that particular operation is
not supported in that configuration.
NAME I/O Operation BW(Read/Write)
9p-cache-none seqread-psync 27(MiB/s)
virtiofs-cache-none seqread-psync 35(MiB/s)
virtiofs-dax-cache-none seqread-psync 245(MiB/s)
9p-cache-none seqread-psync-multi 117(MiB/s)
virtiofs-cache-none seqread-psync-multi 162(MiB/s)
virtiofs-dax-cache-none seqread-psync-multi 894(MiB/s)
9p-cache-none seqread-mmap 24(MiB/s)
virtiofs-cache-none seqread-mmap 0(KiB/s)
virtiofs-dax-cache-none seqread-mmap 168(MiB/s)
9p-cache-none seqread-mmap-multi 115(MiB/s)
virtiofs-cache-none seqread-mmap-multi 0(KiB/s)
virtiofs-dax-cache-none seqread-mmap-multi 614(MiB/s)
9p-cache-none seqread-libaio 26(MiB/s)
virtiofs-cache-none seqread-libaio 139(MiB/s)
virtiofs-dax-cache-none seqread-libaio 160(MiB/s)
9p-cache-none seqread-libaio-multi 129(MiB/s)
virtiofs-cache-none seqread-libaio-multi 142(MiB/s)
virtiofs-dax-cache-none seqread-libaio-multi 577(MiB/s)
9p-cache-none randread-psync 29(MiB/s)
virtiofs-cache-none randread-psync 34(MiB/s)
virtiofs-dax-cache-none randread-psync 256(MiB/s)
9p-cache-none randread-psync-multi 139(MiB/s)
virtiofs-cache-none randread-psync-multi 153(MiB/s)
virtiofs-dax-cache-none randread-psync-multi 245(MiB/s)
9p-cache-none randread-mmap 22(MiB/s)
virtiofs-cache-none randread-mmap 0(KiB/s)
virtiofs-dax-cache-none randread-mmap 162(MiB/s)
9p-cache-none randread-mmap-multi 111(MiB/s)
virtiofs-cache-none randread-mmap-multi 0(KiB/s)
virtiofs-dax-cache-none randread-mmap-multi 215(MiB/s)
9p-cache-none randread-libaio 26(MiB/s)
virtiofs-cache-none randread-libaio 135(MiB/s)
virtiofs-dax-cache-none randread-libaio 157(MiB/s)
9p-cache-none randread-libaio-multi 133(MiB/s)
virtiofs-cache-none randread-libaio-multi 245(MiB/s)
virtiofs-dax-cache-none randread-libaio-multi 163(MiB/s)
9p-cache-none seqwrite-psync 28(MiB/s)
virtiofs-cache-none seqwrite-psync 34(MiB/s)
virtiofs-dax-cache-none seqwrite-psync 203(MiB/s)
9p-cache-none seqwrite-psync-multi 128(MiB/s)
virtiofs-cache-none seqwrite-psync-multi 155(MiB/s)
virtiofs-dax-cache-none seqwrite-psync-multi 717(MiB/s)
9p-cache-none seqwrite-mmap 0(KiB/s)
virtiofs-cache-none seqwrite-mmap 0(KiB/s)
virtiofs-dax-cache-none seqwrite-mmap 165(MiB/s)
9p-cache-none seqwrite-mmap-multi 0(KiB/s)
virtiofs-cache-none seqwrite-mmap-multi 0(KiB/s)
virtiofs-dax-cache-none seqwrite-mmap-multi 511(MiB/s)
9p-cache-none seqwrite-libaio 27(MiB/s)
virtiofs-cache-none seqwrite-libaio 128(MiB/s)
virtiofs-dax-cache-none seqwrite-libaio 141(MiB/s)
9p-cache-none seqwrite-libaio-multi 119(MiB/s)
virtiofs-cache-none seqwrite-libaio-multi 242(MiB/s)
virtiofs-dax-cache-none seqwrite-libaio-multi 505(MiB/s)
9p-cache-none randwrite-psync 27(MiB/s)
virtiofs-cache-none randwrite-psync 34(MiB/s)
virtiofs-dax-cache-none randwrite-psync 189(MiB/s)
9p-cache-none randwrite-psync-multi 137(MiB/s)
virtiofs-cache-none randwrite-psync-multi 150(MiB/s)
virtiofs-dax-cache-none randwrite-psync-multi 233(MiB/s)
9p-cache-none randwrite-mmap 0(KiB/s)
virtiofs-cache-none randwrite-mmap 0(KiB/s)
virtiofs-dax-cache-none randwrite-mmap 120(MiB/s)
9p-cache-none randwrite-mmap-multi 0(KiB/s)
virtiofs-cache-none randwrite-mmap-multi 0(KiB/s)
virtiofs-dax-cache-none randwrite-mmap-multi 200(MiB/s)
9p-cache-none randwrite-libaio 25(MiB/s)
virtiofs-cache-none randwrite-libaio 124(MiB/s)
virtiofs-dax-cache-none randwrite-libaio 131(MiB/s)
9p-cache-none randwrite-libaio-multi 125(MiB/s)
virtiofs-cache-none randwrite-libaio-multi 241(MiB/s)
virtiofs-dax-cache-none randwrite-libaio-multi 163(MiB/s)
- In general virtio-fs seems faster than virtio-9p. Using dax makes it
Right now dax window is 8G and max fio file size is 8G as well (4
files of 2G each). That means everything fits into dax window and no
reclaim is needed. Dax window reclaim logic is slower and if file
size is bigger than dax window size, performance slows down.
Description from previous postings
With the goal of designing something with better performance and local file
system semantics, a bunch of ideas were proposed.
- Use fuse protocol (instead of 9p) for communication between guest
and host. Guest kernel will be fuse client and a fuse server will
run on host to serve the requests.
- For data access inside guest, mmap portion of file in QEMU address
space and guest accesses this memory using dax. That way guest page
cache is bypassed and there is only one copy of data (on host). This
will also enable mmap(MAP_SHARED) between guests.
- For metadata coherency, there is a shared memory region which contains
version number associated with metadata and any guest changing metadata
updates version number and other guests refresh metadata on next
access. This is yet to be implemented.
How virtio-fs differs from existing approaches
The unique idea behind virtio-fs is to take advantage of the co-location
of the virtual machine and hypervisor to avoid communication (vmexits).
DAX allows file contents to be accessed without communication with the
hypervisor. The shared memory region for metadata avoids communication in
the common case where metadata is unchanged.
By replacing expensive communication with cheaper shared memory accesses,
we expect to achieve better performance than approaches based on network
file system protocols. In addition, this also makes it easier to achieve
local file system semantics (coherency).
These techniques are not applicable to network file system protocols since
the communications channel is bypassed by taking advantage of shared memory
on a local machine. This is why we decided to build virtio-fs rather than
focus on 9P or NFS.
Like virtio-9p, different caching modes are supported which determine the
coherency level as well. The “cache=FOO” and “writeback” options control the
level of coherence between the guest and host filesystems.
metadata, data and pathname lookup are not cached in guest. They are always
fetched from host and any changes are immediately pushed to host.
metadata, data and pathname lookup are cached in guest and never expire.
metadata and pathname lookup cache expires after a configured amount of time
(default is 1 second). Data is cached while the file is open (close to open
These options control the writeback strategy. If writeback is disabled,
then normal writes will immediately be synchronized with the host fs. If
writeback is enabled, then writes may be cached in the guest until the file
is closed or an fsync(2) performed. This option has no effect on mmap-ed
writes or writes going through the DAX mechanism.
Miklos Szeredi (2):
fuse: delete dentry if timeout is zero
fuse: Use default_file_splice_read for direct IO
Stefan Hajnoczi (6):
fuse: export fuse_end_request()
fuse: export fuse_len_args()
fuse: export fuse_get_unique()
fuse: extract fuse_fill_super_common()
fuse: add fuse_iqueue_ops callbacks
virtio_fs: add skeleton virtio_fs.ko module
Vivek Goyal (5):
fuse: Export fuse_send_init_request()
Export fuse_dequeue_forget() function
fuse: Separate fuse device allocation and installation in fuse_conn
virtio-fs: Do not provide abort interface in fusectl
init/do_mounts.c: add virtio_fs root fs support
fs/fuse/Kconfig | 11 +
fs/fuse/Makefile | 1 +
fs/fuse/control.c | 4 +-
fs/fuse/cuse.c | 4 +-
fs/fuse/dev.c | 89 ++-
fs/fuse/dir.c | 26 +-
fs/fuse/file.c | 15 +-
fs/fuse/fuse_i.h | 120 +++-
fs/fuse/inode.c | 203 +++---
fs/fuse/virtio_fs.c | 1061 +++++++++++++++++++++++++++++++
fs/splice.c | 3 +-
include/linux/fs.h | 2 +
include/uapi/linux/virtio_fs.h | 41 ++
include/uapi/linux/virtio_ids.h | 1 +
init/do_mounts.c | 10 +
15 files changed, 1462 insertions(+), 129 deletions(-)
create mode 100644 fs/fuse/virtio_fs.c
create mode 100644 include/uapi/linux/virtio_fs.h
More information about the Virtio-fs