tl;dr:
This patch introduces an alternate way of handling the receive
buffers of the spapr-vlan device, resulting in much better
receive performance for the guest.
Full story:
One of our testers recently discovered that the performance of the
spapr-vlan device is very poor compared to other NICs, and that
a simple "ping -i 0.2 -s 65507 someip" in the guest can result
in more than 50% lost ping packets (especially with older guest
kernels < 3.17).
After doing some analysis, it was clear that there is a problem
with the way we handle the receive buffers in spapr_llan.c: The
ibmveth driver of the guest Linux kernel tries to add a lot of
buffers into several buffer pools (with 512, 2048 and 65536 byte
sizes by default, but it can be changed via the entries in the
/sys/devices/vio/1000/pool* directories of the guest). However,
the spapr-vlan device of QEMU only tries to squeeze all receive
buffer descriptors into one single page which has been supplied
by the guest during the H_REGISTER_LOGICAL_LAN call, without
taking care of different buffer sizes. This has two bad effects:
First, only a very limited number of buffer descriptors is accepted
at all. Second, we also hand 64k buffers to the guest even if
the 2k buffers would fit better - and this results in dropped packets
in the IP layer of the guest since too much skbuf memory is used.
Though it seems at a first glance like PAPR says that we should store
the receive buffer descriptors in the page that is supplied during
the H_REGISTER_LOGICAL_LAN call, chapter 16.4.1.2 in the LoPAPR spec
declares that "the contents of these descriptors are architecturally
opaque, none of these descriptors are manipulated by code above
the architected interfaces". That means we don't have to store
the RX buffer descriptors in this page, but can also manage the
receive buffers at the hypervisor level only. This is now what we
are doing here: Introducing proper RX buffer pools which are also
sorted by size of the buffers, so we can hand out a buffer with
the best fitting size when a packet has been received.
To avoid problems with migration from/to older version of QEMU,
the old behavior is also retained and enabled by default. The new
buffer management has to be enabled via a new "use-rx-buffer-pools"
property.
Now with the new buffer pool management enabled, the problem with
"ping -s 65507" is fixed for me, and the throughput of a simple
test with wget increases from creeping 3MB/s up to 20MB/s!
Signed-off-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
QEMU README
===========
QEMU is a generic and open source machine & userspace emulator and
virtualizer.
QEMU is capable of emulating a complete machine in software without any
need for hardware virtualization support. By using dynamic translation,
it achieves very good performance. QEMU can also integrate with the Xen
and KVM hypervisors to provide emulated hardware while allowing the
hypervisor to manage the CPU. With hypervisor support, QEMU can achieve
near native performance for CPUs. When QEMU emulates CPUs directly it is
capable of running operating systems made for one machine (e.g. an ARMv7
board) on a different machine (e.g. an x86_64 PC board).
QEMU is also capable of providing userspace API virtualization for Linux
and BSD kernel interfaces. This allows binaries compiled against one
architecture ABI (e.g. the Linux PPC64 ABI) to be run on a host using a
different architecture ABI (e.g. the Linux x86_64 ABI). This does not
involve any hardware emulation, simply CPU and syscall emulation.
QEMU aims to fit into a variety of use cases. It can be invoked directly
by users wishing to have full control over its behaviour and settings.
It also aims to facilitate integration into higher level management
layers, by providing a stable command line interface and monitor API.
It is commonly invoked indirectly via the libvirt library when using
open source applications such as oVirt, OpenStack and virt-manager.
QEMU as a whole is released under the GNU General Public License,
version 2. For full licensing details, consult the LICENSE file.
Building
========
QEMU is multi-platform software intended to be buildable on all modern
Linux platforms, OS-X, Win32 (via the Mingw64 toolchain) and a variety
of other UNIX targets. The simple steps to build QEMU are:
mkdir build
cd build
../configure
make
Complete details of the process for building and configuring QEMU for
all supported host platforms can be found in the qemu-tech.html file.
Additional information can also be found online via the QEMU website:
http://qemu-project.org/Hosts/Linux
http://qemu-project.org/Hosts/W32
Submitting patches
==================
The QEMU source code is maintained under the GIT version control system.
git clone git://git.qemu-project.org/qemu.git
When submitting patches, the preferred approach is to use 'git
format-patch' and/or 'git send-email' to format & send the mail to the
qemu-devel@nongnu.org mailing list. All patches submitted must contain
a 'Signed-off-by' line from the author. Patches should follow the
guidelines set out in the HACKING and CODING_STYLE files.
Additional information on submitting patches can be found online via
the QEMU website
http://qemu-project.org/Contribute/SubmitAPatch
http://qemu-project.org/Contribute/TrivialPatches
Bug reporting
=============
The QEMU project uses Launchpad as its primary upstream bug tracker. Bugs
found when running code built from QEMU git or upstream released sources
should be reported via:
https://bugs.launchpad.net/qemu/
If using QEMU via an operating system vendor pre-built binary package, it
is preferable to report bugs to the vendor's own bug tracker first. If
the bug is also known to affect latest upstream code, it can also be
reported via launchpad.
For additional information on bug reporting consult:
http://qemu-project.org/Contribute/ReportABug
Contact
=======
The QEMU community can be contacted in a number of ways, with the two
main methods being email and IRC
- qemu-devel@nongnu.org
http://lists.nongnu.org/mailman/listinfo/qemu-devel
- #qemu on irc.oftc.net
Information on additional methods of contacting the community can be
found online via the QEMU website:
http://qemu-project.org/Contribute/StartHere
-- End