Re: how to benchmark schedulers

[Qu Wenruo]

On 2017年08月08日 21:19, Janos Toth F. wrote:

I think you should consider using Linux 4.12 which has bfq (bfq-mq)
for blk-mq. So, you don't need out-of-tree BFQ patches if you switch
to blk-mq (but now you are free to do so even if you have HDDs or SSDs
which benefit from software schedulers since you have some multi-queue
schedulers for them). Just make sure to enable blk-mq (has to be a
boot parameter or build-time choice) in order to gain access to
bfq-mq. And remember that bfq-mq has to be activated manually (the
build-time choice for a default scheduler is not valid for multi-queue
schedulers, you will default to "none" which is effectively the new
"no-op").
Note: there is only one BFQ in 4.12 and it's bfq-mq which runs under
the name of simply BFQ (not bfq-mq, I only used that name to make it
clear that BFQ in 4.12 is a multi-queue version of BFQ).

I always wondered if Btrfs makes any use of FUA if it's enabled for
compatible SATA devices (it's disabled by default because there are
some drives with faulty firmware).


At least, if using blktrace, you could find that superblock write of 
btrfs is submitting bio using FWFSM (Flush, Write, FUA, Sync, Meta).


So for compatible SATA device, it is possible to use FUA.
It's block driver to emulate FUA by addressing Write then Flush for 
incompatible disk.


However, in my previous test with FUA compatible HDD (you could check 
kernel "Write cache:" line), it turns out that at least for DB workload, 
the performance improvement is hard to notice and sometimes even reduce 
the performance.


Maybe it's related to the workload (PostgreSQL), but FUA support may or 
may not affect performance as you expected.


Thanks,
Qu


I also wonder if RAID10 is any better (or actually worse?) for
metadata (and system) chunks than RAID1.

On Tue, Aug 8, 2017 at 1:59 PM, Bernhard Landauer  wrote:

Hello everyone

I am looking for a way to test different available schedulers with Manjaro's
bfq-patched kernels on sytems with both SSD and spinning drives. Since
phoronix-test-suite apparently is currently useless for this task due to its
bad config for bfq I am looking for alternatives. Do you have any
suggestions for me?
Thank you.

kind regards
Bernhard Landauer
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html


--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: how to benchmark schedulers

[Bernhard Landauer]

On 08.08.2017 15:19, Janos Toth F. wrote:


I think you should consider using Linux 4.12 which has bfq (bfq-mq)
for blk-mq. So, you don't need out-of-tree BFQ patches if you switch
to blk-mq (but now you are free to do so even if you have HDDs or SSDs
which benefit from software schedulers since you have some multi-queue
schedulers for them). Just make sure to enable blk-mq (has to be a
boot parameter or build-time choice) in order to gain access to
bfq-mq. And remember that bfq-mq has to be activated manually (the
build-time choice for a default scheduler is not valid for multi-queue
schedulers, you will default to "none" which is effectively the new
"no-op").
Note: there is only one BFQ in 4.12 and it's bfq-mq which runs under
the name of simply BFQ (not bfq-mq, I only used that name to make it
clear that BFQ in 4.12 is a multi-queue version of BFQ).

I always wondered if Btrfs makes any use of FUA if it's enabled for
compatible SATA devices (it's disabled by default because there are
some drives with faulty firmware).

I also wonder if RAID10 is any better (or actually worse?) for
metadata (and system) chunks than RAID1.

On Tue, Aug 8, 2017 at 1:59 PM, Bernhard Landauer  wrote:

Hello everyone

I am looking for a way to test different available schedulers with Manjaro's
bfq-patched kernels on sytems with both SSD and spinning drives. Since
phoronix-test-suite apparently is currently useless for this task due to its
bad config for bfq I am looking for alternatives. Do you have any
suggestions for me?
Thank you.

kind regards
Bernhard Landauer
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Thank you for your reply, Janos!
In fact what I am trying to do is evaluating which kernel and which 
schedulers are delivering the best results for me and I would like to 
test/benchmark the differences.

So my question is about available tests I could run with different setups.
Please let me know if you know of any.

regards
Bernhard
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: how to benchmark schedulers

[Janos Toth F.]

I think you should consider using Linux 4.12 which has bfq (bfq-mq)
for blk-mq. So, you don't need out-of-tree BFQ patches if you switch
to blk-mq (but now you are free to do so even if you have HDDs or SSDs
which benefit from software schedulers since you have some multi-queue
schedulers for them). Just make sure to enable blk-mq (has to be a
boot parameter or build-time choice) in order to gain access to
bfq-mq. And remember that bfq-mq has to be activated manually (the
build-time choice for a default scheduler is not valid for multi-queue
schedulers, you will default to "none" which is effectively the new
"no-op").
Note: there is only one BFQ in 4.12 and it's bfq-mq which runs under
the name of simply BFQ (not bfq-mq, I only used that name to make it
clear that BFQ in 4.12 is a multi-queue version of BFQ).

I always wondered if Btrfs makes any use of FUA if it's enabled for
compatible SATA devices (it's disabled by default because there are
some drives with faulty firmware).

I also wonder if RAID10 is any better (or actually worse?) for
metadata (and system) chunks than RAID1.

On Tue, Aug 8, 2017 at 1:59 PM, Bernhard Landauer  wrote:
> Hello everyone
>
> I am looking for a way to test different available schedulers with Manjaro's
> bfq-patched kernels on sytems with both SSD and spinning drives. Since
> phoronix-test-suite apparently is currently useless for this task due to its
> bad config for bfq I am looking for alternatives. Do you have any
> suggestions for me?
> Thank you.
>
> kind regards
> Bernhard Landauer
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majord...@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

how to benchmark schedulers

[Bernhard Landauer]

Hello everyone

I am looking for a way to test different available schedulers with 
Manjaro's bfq-patched kernels on sytems with both SSD and spinning 
drives. Since phoronix-test-suite apparently is currently useless for 
this task due to its bad config for bfq I am looking for alternatives. 
Do you have any suggestions for me?

Thank you.

kind regards
Bernhard Landauer
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html