Felix Fietkau writes:
> On 2017-11-08 07:18, John Crispin wrote:
> It's just the I/O scheduler, not the CPU one. It will have zero impact
> on typical router workloads. The only thing that might get slower is
> heavy multi-threaded disk I/O, which is probably an extremely rare
> occurence on LEDE
On Wed, 2017-11-08 at 10:27 +0100, Felix Fietkau wrote:
> On 2017-11-08 07:18, John Crispin wrote:
> > j
> >
> >
> > On 07/11/17 19:41, Rosen Penev wrote:
> > > most users don't have multithreaded workloads though.
> > >
> > > On Mon, Nov 6, 2017 at 4:43 PM, Dave Taht wrote:
> > > > I happen to
On 2017-11-08 07:18, John Crispin wrote:
> j
>
>
> On 07/11/17 19:41, Rosen Penev wrote:
>> most users don't have multithreaded workloads though.
>>
>> On Mon, Nov 6, 2017 at 4:43 PM, Dave Taht wrote:
>>> I happen to like deadline schedulers, and at least from a kernel
>>> perspective, we have a
j
On 07/11/17 19:41, Rosen Penev wrote:
most users don't have multithreaded workloads though.
On Mon, Nov 6, 2017 at 4:43 PM, Dave Taht wrote:
I happen to like deadline schedulers, and at least from a kernel
perspective, we have a very large set of multithreaded workloads.
So I would not ma
most users don't have multithreaded workloads though.
On Mon, Nov 6, 2017 at 4:43 PM, Dave Taht wrote:
>
> I happen to like deadline schedulers, and at least from a kernel
> perspective, we have a very large set of multithreaded workloads.
>
> So I would not make this change without a very seriou
I happen to like deadline schedulers, and at least from a kernel
perspective, we have a very large set of multithreaded workloads.
So I would not make this change without a very serious set of benchmarks
under load showing it makes a difference.
___
Le
I haven't made any benchmarks, no. The router that I use is the Turris
Omnia which has pretty fast eMMC instead of NOR, so they probably
wouldn't be useful to compare to the NOR based stuff.
LEDE tends not to write to flash but does write to RAM, which I think
is still impacted by the choice of sc
No-op uses less CPU compared to deadline. Important since LEDE runs mainly on
routers. Getting rid of deadline saved 156 bytes on my vmlinux file.
Tests conducted by Phoronix on the 4.12 kernel indicate that Deadline is
slightly slower than No-op, except with multi-threaded workloads.
https://w
