Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 8 Sep 2013 17:56:15 +0800
From:      zhixiang yu <zxyu.core@gmail.com>
To:        Alexander Motin <mav@freebsd.org>
Cc:        soc-status@freebsd.org
Subject:   Re: Weekly report of the passed two weeks
Message-ID:  <CADr9iNv=pDQO1tSNM%2BCuv9YfTpmubnjkAEcFZg1cTFhvzaUuuQ@mail.gmail.com>
In-Reply-To: <522B72D2.7070906@FreeBSD.org>
References:  <CADr9iNvmm-aX8SO3DQshzaNTS204%2Bep9gvGiOaymw4QMp=2-sg@mail.gmail.com> <522B72D2.7070906@FreeBSD.org>

next in thread | previous in thread | raw e-mail | index | archive | help
Hi Alexander,

On Sun, Sep 8, 2013 at 2:39 AM, Alexander Motin <mav@freebsd.org> wrote:
> Hi.
>
> For stress test I would try to create ZFS pool on the device, disable ARC
> caching and run some heavy filesystem benchmark. ZFS controls disk subsystem
> sanity on every read and should report you any data inconsistency in `zpool
> status` output.

It seems that 'zpool status' is very convenient. I'm not familiar with
ZFS and never tried it before. So could you please give one detailed
steps to do so?

>
>
>> You can try it yourself ;)
>>
>> The ATAPI commands' support is on-going. Maybe finished by next
>> weekend. I won't need the same much time as ATA because I'm more
>> familiar with read/write emulation.
>
>
> While they are not recommended for use by AHCI specification, and may be not
> required to boot FreeBSD guest, I would still implement ATA PIO commands too
> because they are still mandatory for any ATA device. Fortunately since AHCI
> hides PIO transfer details it should be easy.
>
> Also, looking on the code, I think that sending any unsupported command to
> the device now will cause command timeout, while it should return ABORT
> error status instead. And in general make sure that you implemented all
> commands support for which you announced in IDENTIFY command output.
>
> Also I think it would be good to make FLUSHCACHE commands not only wait for
> write completion, but really call fsync() on the underlying file. That may
> be important for client filesystem consistency in case of host system crash.

Good idea. Will do

Thanks in advance,
Jisheng



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CADr9iNv=pDQO1tSNM%2BCuv9YfTpmubnjkAEcFZg1cTFhvzaUuuQ>