> If there is transactional data the transaction monitor and/or the
> database would typically worry about data being written to persistent
> data store so that in case of an outage you have roll-back or
> roll-forward capabilities.

I think we're starting from different assumptions. I start from the
assumption that the majority of Linux users are storing data on normal
filesystems rather than in database applications.

Technically one could argue that a journaling filesystem may play this
role, but I've seen too many cases where a Linux system does not recover
from a snapshot-based backup.

> If there is "on the fly" generated data that nobody worries about
> if being synced to disk at a particular point in time, a Linux
> crash would certainly cause data loss for data not flushed to
> disk yet.
> If anyone worries about such possible data loss, you better talk to
> your application provider articulating your concern about its data
> hardening policy for critical data.

I doubt we'll get changes in ext2/ext3/reiserfs/xfs/etc. All use the
buffering in memory technique extensively for performance reasons on the
Intel side, and we're not going to change that on 390.

> If the customer uses GDPS/XRC for data replication of the Linux
> environment (VM doesn't time stamp its data, though it honours Linux
> writing timestamps) or alternatively GDPS/PPRC with or without
> HyperSwap for both, Linux and z/VM they probably get as close as ever
> possible, minimizing any critical data loss you probably can. And
> in case of transactional data you typically also have full data
> consistency assurance.

For transactional applications, I might agree. However, looking at the
dozen or so running Linux instances I have easy access to here, I have
approximately 100-200M of unwritten data in storage that I have no
guarantee that it has been flushed to disk. The only way to be *certain*
that that data is committed to a recoverable backup is to run a client
inside those Linux machines, so that the backup application is aware of
the *actual* state of the data and represents it correctly in the
backup.

Another good reason for IBM SSD to publish the control interfaces...8-)

> Therewith depending on your data loss objective in case of a
> disaster scenario a backup like the one TSM usually delivers might
> be good enough, or more sophisticated concurrent data replication
> DR setup like the one delivered by like GDPS/XRC or GDPS/PPRC
> is mandatory - if you have the opportunity integrating Linux into
> your z/OS BR setup that GDPS provides support for.

I'm not saying that GDPS is a bad idea -- if your applications can take
advantage of it, great. I simply maintain that it does not produce
reliable backups for a large class of users.





>
> Best regards,
> Ingo
>
> --
> Ingo Adlung,
> zSeries Linux and Virtualization Architecture
> mail: [EMAIL PROTECTED] - phone: +49-7031-16-4263
>
> Linux on 390 Port <[email protected]> wrote on
> 04.05.2005 16:36:30:
>
> > > I
> > > was thinking
> > > to do a "Sync and Drop" process on the Linux volumes , the
> > > LPAR would remain
> > > up during this process. Has anyone else already looked at
> > > this ?
> >
> > You will not get a usable backup. This process does not take into
> > account the data cached in memory.
> >
> > > Any ideas what the
> > > change would
> > > be if z/VM was in place .
> >
> > You still can't use Sync and Drop reliably. You need a
> backup solution
> > that is aware of what's going on inside the Linux system,
> eg something
> > with a Linux client like Bacula or TSM.
> >
> >
> ----------------------------------------------------------------------
> > For LINUX-390 subscribe / signoff / archive access instructions,
> > send email to [EMAIL PROTECTED] with the message: INFO
> LINUX-390 or
> visit
> > http://www.marist.edu/htbin/wlvindex?LINUX-390
>
> ----------------------------------------------------------------------
> For LINUX-390 subscribe / signoff / archive access instructions,
> send email to [EMAIL PROTECTED] with the message: INFO
> LINUX-390 or visit
> http://www.marist.edu/htbin/wlvindex?LINUX-390
>
>
>  ** ACCEPT: CRM114 PASS Markovian Matcher **
> CLASSIFY succeeds; success probability: 1.0000  pR: 212.0310
> Best match to file #0 (nonspam.css) prob: 1.0000  pR: 212.0310
> Total features in input file: 13344
> #0 (nonspam.css): features: 3775612, hits: 3219461, prob:
> 1.00e+00, pR: 212.03
> #1 (spam.css): features: 3788298, hits: 2820935, prob:
> 9.31e-213, pR: -212.03
>
>

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to