samba - Sep 2016 - Multichannel working at half speed with 2x NICs

Similar to another thread that was posted recently (archive:
https://groups.google.com/forum/#!topic/linux.samba/sRvmCb25U_s) I'm trying
to get multichannel working with a virtual Linux server running version
4.4.5.

Output of 'uname -a': "Linux filehost 4.4.0-38-generic #57-Ubuntu
SMP Tue
Sep 6 15:42:33 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux"

I have two VirtIO NICs on my server that I've confirmed works at at least
1Gbps
(via iperf) to my Windows 10 client which also has 2 1Gbps NICs. Since virtio
interfaces dont seem to report link speeds, I've changed my smb.conf to
reflect this as such:

"interfaces = 192.168.1.120;speed=1000000000
192.168.1.118;speed=1000000000"

I've also tried with and without "aio read size = 1" and "aio
write size 1", but that didn't change anything. In addition, I've
tried all the
suggestions in thread linked above and all have been unsuccessful.

I've verified that all my drives on both the server and client are capable
of sustaining read/writes of well over 250MB/s, so I don't think this is a
disk bottleneck issue.

What happens during a file transfer is that both interfaces are only
utilized at around 50%, so they both combine to give me ~1Gbps, when they
really should be working at 100% each for ~2Gbps total.

What can I do to troubleshoot this?

Thanks in advance for any help.

On Wed, Sep 28, 2016 at 05:56:24AM +0000, Kanchana Marasinghe via samba
wrote:
> Similar to another thread that was posted recently (archive:
> https://groups.google.com/forum/#!topic/linux.samba/sRvmCb25U_s) I'm
trying
> to get multichannel working with a virtual Linux server running version
> 4.4.5.
> 
> Output of 'uname -a': "Linux filehost 4.4.0-38-generic
#57-Ubuntu SMP Tue
> Sep 6 15:42:33 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux"
> 
> I have two VirtIO NICs on my server that I've confirmed works at at
least 1Gbps
> (via iperf) to my Windows 10 client which also has 2 1Gbps NICs. Since
virtio
> interfaces dont seem to report link speeds, I've changed my smb.conf to
> reflect this as such:
> 
> "interfaces = 192.168.1.120;speed=1000000000
192.168.1.118;speed=1000000000"
> 
> I've also tried with and without "aio read size = 1" and
"aio write size > 1", but that didn't change anything. In
addition, I've tried all the
> suggestions in thread linked above and all have been unsuccessful.
> 
> I've verified that all my drives on both the server and client are
capable
> of sustaining read/writes of well over 250MB/s, so I don't think this
is a
> disk bottleneck issue.
> 
> What happens during a file transfer is that both interfaces are only
> utilized at around 50%, so they both combine to give me ~1Gbps, when they
> really should be working at 100% each for ~2Gbps total.
> 
> What can I do to troubleshoot this?
strace the connected smbd to ensure it's using the pthreadpool.

Hi Jeremy,

Is it safe to assume it's just "strace -p <pid>", for the
running samba
daemon?
Regards
Kanchana

On Wed, Sep 28, 2016, 1:00 PM Jeremy Allison <jra at samba.org> wrote:
> On Wed, Sep 28, 2016 at 05:56:24AM +0000, Kanchana Marasinghe via samba
> wrote:
> > Similar to another thread that was posted recently (archive:
> > https://groups.google.com/forum/#!topic/linux.samba/sRvmCb25U_s)
I'm
> trying
> > to get multichannel working with a virtual Linux server running
version
> > 4.4.5.
> >
> > Output of 'uname -a': "Linux filehost 4.4.0-38-generic
#57-Ubuntu SMP Tue
> > Sep 6 15:42:33 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux"
> >
> > I have two VirtIO NICs on my server that I've confirmed works at
at
> least 1Gbps
> > (via iperf) to my Windows 10 client which also has 2 1Gbps NICs. Since
> virtio
> > interfaces dont seem to report link speeds, I've changed my
smb.conf to
> > reflect this as such:
> >
> > "interfaces = 192.168.1.120;speed=1000000000
> 192.168.1.118;speed=1000000000"
> >
> > I've also tried with and without "aio read size = 1" and
"aio write size
> > > 1", but that didn't change anything. In addition,
I've tried all the
> > suggestions in thread linked above and all have been unsuccessful.
> >
> > I've verified that all my drives on both the server and client are
> capable
> > of sustaining read/writes of well over 250MB/s, so I don't think
this is
> a
> > disk bottleneck issue.
> >
> > What happens during a file transfer is that both interfaces are only
> > utilized at around 50%, so they both combine to give me ~1Gbps, when
they
> > really should be working at 100% each for ~2Gbps total.
> >
> > What can I do to troubleshoot this?
>
> strace the connected smbd to ensure it's using the pthreadpool.
>