Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 27 Mar 2015 10:25:05 +1030
From:      "O'Connor, Daniel" <darius@dons.net.au>
To:        d@delphij.net
Cc:        freebsd-hackers@freebsd.org, Wojciech Puchar <wojtek@puchar.net>
Subject:   Re: Seagate Archive HDD
Message-ID:  <4F65B315-5FFE-4184-91FD-C05A40E0A26E@dons.net.au>
In-Reply-To: <55148E42.80708@delphij.net>
References:  <alpine.BSF.2.20.1503261124380.1417@laptop.wojtek.intra> <55148E42.80708@delphij.net>

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

> On 27 Mar 2015, at 09:24, Xin Li <delphij@delphij.net> wrote:
> On 03/26/15 03:26, Wojciech Puchar wrote:
>> http://www.storagereview.com/seagate_archive_hdd_review_8tb
>>=20
>> i want to buy 2 such drives for backup server.
>>=20
>> This drives use shingled recording.
>>=20
>> Are anyone using them and can confirm they are compatible on
>> software level with other disks? I understand average random write
>> time would be 5-10 times slower than normal drive because of the
>> need of rewrite few full tracks worth of data, but otherwise will
>> then be compatible and can i use it as usual?
>=20
> My understanding is that the SMR drives are actually different class
> of storage device.
>=20
> The "drive managed" drives as shipped now tries to emulate normal hard
> drive's behavior but they present unique risks: for instance, a
> rewrite of a small block may end up in a read-modify-write of a much
> larger area, so we must refrain from doing such operations for
> critical file system data structure, probably by reorganizing the
> on-disk format to satisfy the need.

I don't think this is necessarily true - I watched a very informative =
presentation about SMR drives - =
https://www.usenix.org/conference/fast15/technical-sessions/presentation/a=
ghayev

The drive has many regions with guard bands so it doesn't have to =
rewrite the entire disk for certain writes (it may have to rewrite a =
whole band though)

It also has a log section which it writes to and then back fills into =
the shingled area later.

The upshot is that you get very weird latencies, but generally writing =
sequentially is OK. Random read is OK, random write is utterly abysmal =
(duh).

One thing I would like to know is if the drive supports TRIM so it can =
avoid rewriting a band or not.

--
Daniel O'Connor
"The nice thing about standards is that there
are so many of them to choose from."
 -- Andrew Tanenbaum
GPG Fingerprint - 5596 B766 97C0 0E94 4347 295E E593 DC20 7B3F CE8C




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4F65B315-5FFE-4184-91FD-C05A40E0A26E>