Roy Schestowitz wrote:
> __/ [ BearItAll ] on Wednesday 29 March 2006 09:37 \__
>> Roy Schestowitz wrote:
>>> Linuxworld: Linux continues Unix bloodbath
>>> ,----[ Quote ]
>>> | Even the most mission-critical, Unix-based central business systems
>>> | are being replaced by Linux running on commodity hardware, according
>>> | to delegates at Linuxworld 2006 conference and expo in Sydney
>>> | yesterday.
>> That is basically what I recommend to companies. Always mirror/cluster,
>> but don't worry about using cheaper machines to save costs. I always have
>> to remind them that a mirror isn't a backup (you wouldn't believe the
>> number of companies I visit where they assume a mirror means a backup
>> isn't necessary).
>> All my servers here are just cheap off the shelf machines, except one, a
>> Dell Poweredge which only has the job of a mirror and backup control,
>> because I can't trust it to be a live server. Typical isn't it, the most
>> expensive machine I have and it will never actually do the job of a
> Mirror != Backup? What's deficient in the following, for example?
> ,----[ Code ]
> | # Locally
> | nice scp -r ~/.[0-z]* ~/* roy@home:/home/roy/Main/BU/Baine-`date
> | +%Y-%m-%d`
> | # On remote side
> | mkdir /media/SEA_DISK/Home/`date +%Y-%m-%d`
> | tar -cf - /home/roy/Main/BU|split -b 1000m - /media/SEA_DISK/Home/`date
> | +%Y-%m-%d`/Baine-`date +%Y-%m-%d`.tar.
> I am not confuting your opinion. Rather, I am asking for your kind
> advice... I have mirrors on 3 hard-drives and the O/S is restorable from
> the CD/s.
You are doing dated backups.
But I come across system where they have a mirror, either a drive mirror or
machine-machine mirror. But have no way to go back to previous versions of
This current company I am sat in is a fairly typical one. I have the live
Linux servers, each mirrored, but then I added to that by making each
machine capable of standing in for any other. Basically a cluster.
But the auditors are in for the yearly review, they can point at anything
and we must be able to bring up the production path information for it.
They can also pick out a date at random from the beginning of the company,
and we have to produce the detailed paperwork,
>From the backups I can pull in the sate of the company in any month of any
year. The quantity of data could be kept live, but would slow down the
But also for normal documents, we have some that are only used monthly or
yearly. So for example if a mistake was made when a spreadsheet was used
last year, you know how MS documents keep a placeholder inside the document
itself which can make the copy on the server useless if the document is not
closed properly, I can go back to that document prior to its last edit.
All of this is semi-automatic, I prefer to keep some manual input to backups
so that i don't become complacent. But all is done with standard UNIX/Linux
commands in scripts.
I test the backups regularly, though the temptation to be complacent with
that is strong because other than the occasional tape drive error it has
all been faultless for years.
I am lucky because very early in my UNIX carreer I almost lost ALL of a
companies data, it was very likely that it would have cost the company too
in that particular situation, and it was only through lazyness on my part.
They were two days (over a weekend) when I was trying all kinds of things
to get it back and dreading the monday morning when I would have to tell
them what I had done. But a chance remark by someone helped me out and I
managed to get everything back.
The shock to my system has made me very over-the-top as far as backups are
concerned ever since.
But I get that nervous feeling back when I visit a company who isn't
properly backing up. This company I am in at the moment, on my first day I
looked at the backup system (they only had one UNIX server at the time), I
also tried a partial recovery off the tapes. I found that the tape drive
had actually stopped working properly around one year before, but nobody
knew so they just kept rotating the tapes. Which meant that the only backup
they had was a year old.
I wish I could show you the white faces I saw when I told the directors in
the first meeting we had after I arrived.
As for mirrors. These are different. Mirrors allow you to get back to a
working system very quickly, but do not necessarily carry historical
backups of data. The speed of mirroring depends on the servers use, from
instant mirroring to onver night mirroring.
A place such as a bank needs instant, because you don't want to lose a
single transaction. For normal file services it can be wiser to mirror
overnight, effectively making the mirror the first line backup for
documents deleted or damaged today, at the same time protecting your main
system from mirroring anything bad. An example of that is one that happened
here not long ago, someone imported about 4000 records into one of the
customer databases, but they hadn't properly matched the fields. The quotes
side of the database instantly failed. It didn't take long at all to pull
back the data off the mirror, had the mirror been instant, I would have had
to pull it off a tape instead.