Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 21 Jun 2012 18:02:52 +0200
From:      Matthias Gamsjager <mgamsjager@gmail.com>
Cc:        FreeBSD Questions <freebsd-questions@freebsd.org>
Subject:   Re: Is ZFS production ready?
Message-ID:  <AA485AA5-F3F8-4D02-8F59-05AC76172C40@gmail.com>
In-Reply-To: <alpine.BSF.2.00.1206211707570.3361@wojtek.tensor.gdynia.pl>
References:  <4FE2CE38.9000100@gmail.com> <CAPj0R5Kmi-%2BdJ7mPvTrTAoS8O983svOyR2WyK2_v1Cr07dSS_A@mail.gmail.com> <alpine.BSF.2.00.1206211413140.2263@wojtek.tensor.gdynia.pl> <CA%2BD9QhuQ%2BbxKW9%2BdX%2BzS9mErwz8JSkV2G7qL0KfB8BH_LGJAgA@mail.gmail.com> <alpine.BSF.2.00.1206211539230.2903@wojtek.tensor.gdynia.pl> <CA%2BD9QhvR_eKtVxdKcaMyOS7tLw_AOHKgUy3o7mJn2b=chMA0Xw@mail.gmail.com> <alpine.BSF.2.00.1206211619250.3092@wojtek.tensor.gdynia.pl> <CA%2BD9QhvwKZm7heoe7tpfhYCJvkknw_HC7aFjCu%2B-1xYQBmV6ng@mail.gmail.com> <alpine.BSF.2.00.1206211644350.3170@wojtek.tensor.gdynia.pl> <CA%2BD9QhsyOh34SghWzQPpnTig%2BUmSEO2VP7jfPxTXs9zW9Uakeg@mail.gmail.com> <alpine.BSF.2.00.1206211707570.3361@wojtek.tensor.gdynia.pl>

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


On 21 jun. 2012, at 17:15, Wojciech Puchar <wojtek@wojtek.tensor.gdynia.pl> w=
rote:

>>=20
>> I do understand your setup but I dont have too agree that it is a good
>=20
> so i would repeat my question.
> Assume you have 48 disks, in mirrored configuration (24 mirrors) and 480 u=
sers with their data on them.
>=20
> Your solution with ZFS - ZFS crashes or you get double disk failure.
> Assuming the latter by average one per 24 file (randomly chosen) is destro=
yed which - in practice and limited time, means everything destroyed. Actual=
ly more than one per 24 - large files can be spread over.
>=20
> Your solution with UFS - better as there is fsck which slowly but successf=
ully repairs problem. with double disk failure - the same!
>=20
>=20
> You restore everything from backup (i assume you have one). This takes lik=
e a day or more, one or two complete work days lost+all users in practice lo=
st everything  since last backup.
>=20
> My solution with UFS - fsck in case of failure work in parallel on 24 disk=
s so not that long. double disk failure means losing data of 1/24 users.
>=20
> every one per 24 user cannot work, others work and i without any stress do=
 recover this 1/24 of users data from backup after putting replacement disks=
.
>=20
> 1/24 of users lost data since last backup, and some hours of time.
>=20
>=20
> Even assuming ZFS is perfect then we both have problems as often, but my p=
roblems are 1/24 as severe as yours.
>=20
>=20
> Just don't ask me for help when unhappy users will want to cut off your he=
ad.
>=20
>>> And you've never seen me, yet i still exist.
>>>=20
>>=20
>> Really? that's you anwser to my question. The most childish answer I coul=
d
>=20
> stupid answer to stupid question.
> You never seen - but they do happens.

In other topic you hammerd on  fact and if someone ask you to deliver them i=
ts a stupid question.=20

And about the dram error. I really hope you do use ecc memory in production w=
hich renders your scenario invalide. And even then its a claim made by you s=
ome random dude on a list.=20

Without proper test scenario and documentation such claims are just useless.=
=20

And a proper layout zfs will withstand a double disk failure with zero downt=
ime...where younhave to tell your customer they just lost a day work=



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AA485AA5-F3F8-4D02-8F59-05AC76172C40>