cancel
Showing results for 
Search instead for 
Did you mean: 

Synthetic Backups with Differentials?

Simon_Gadsby
Level 4
Partner
Hi,

Is it possible to use differential backups with Synthentic Backups instead of Incrementals?
If so, can I overwrite all differentials except the last one and still get a successful Synthetic Backup?

And is it possible to synthetically generate *Differential* backups from a set of Incremental backups? ie. some kind of cumulative incremental?

I want to know whether it is possible to do something like this:
* monthly baseline backup
* weekly synthetics
* Incrementals during the day
* rollup incrementals written to a single tape nightly

I guess a standard duplicate job in the policy will write the daily incrementals to a single tape ok... that might work. But then I have to keep all the incrementals around for a month until the next baseline. It would be nice to regularly roll them up with a differential.

Thoughts?

Thanks,
Simon.

8 REPLIES 8

Simon_Gadsby
Level 4
Partner
I realised, the problem with incrementals is that I will need ALL of them available to restore data from a specified date. Assuming they are no longer available on disk when I need to restore something, this could conceivably mean that I need every single tape used from the prior baseline backup to the point of restore. I would prefer to use the previous baseline + 1 tape containing a differential (or rollup incremental) in order to restore the data, and reuse the intervening tapes.

Colin_Weaver
Moderator
Moderator
Employee Accredited Certified
The way Synthetic works is
1) Baseline runs
2) One or more incrementals runs
3) Synthetic runs that needs access to the basline and the incrementals
4) One or more incrementals runs
5) Synthetic runs that needs access to the previous synthetic and the incrementals since that synthetic ran.
6) repeat from step 4
Restore can be done from last synthetic and all incrementals since last synthetic

The way Incrementals without synthetic work is
1) Run Full backup once a week (which in theory takes longer than the synthetic process, although in practice depending on your file strructure synthetic backups can take extended time to index the files they backup in order to build the full backup)
2) Run incrementals all week
Restore can be done from last full and all incrementals since last full (potentially lots of media)

The way differentials work without synthetic is
1) Run Full backup once a week
2) Run Differentials all week
Restore can be done from last full and last  differential (2 sets of media)

If you are considering Differentials then there is really no reason to use Synthetic anyway as a Differential will need the last full and the last Differential, so run a full once a week and differentials all week and only need 2 tapes for restore - which is probably why Synthetics don't currently use Differentials.

Also note all of the strategies need the full (or synthetic to make a full) running periodically - you should not rely on Incrementals or differentials on their own for an extended period (although it can be longer than 1 week depending on your strategy

Simon_Gadsby
Level 4
Partner
Thanks heaps for clarifying Colin. I have been planning on Synthetic backups for the primary benefit of a reduced backup window, however I might not be able to make it work.

It seems that Synthetic Backups will work from incrementals only.

One thing to clarify: if I write my baseline to disk & synthetic to tape, can I use the same baseline on disk for a future synthetic? I am assuming yes, however I will need all incrementals from that base image onwards in order to produce a future synthetic.

So it seems to me that I cannot use differentials at all with this strategy. I suppose I might be able to include differentials in the same policy, but they would have to be based on modified time instead of archive bit, & would run from the baseline. If I do monthly baselines then the differentials will become too large.

That's a pain.. there appears to be 2 options for using synthetic backup:
1) write full synthetics each night (1 backup set required for restore)
2) keep all incrementals available (all backup sets since last baseline required for restore)

Neither works for me. :(

So I might have to give up the benefits of synthetic & just do full backups weekly as you suggest. Very frustrating. I also notice that the benefits of True Image Restore only come with incremental backups as well, and not with differentials.

Colin_Weaver
Moderator
Moderator
Employee Accredited Certified

The baseline is only used once by the first synthetic - subsequent synthetics must have access to the last syntehtic and cannot use the baseline - the normal configuration is do all Baseline , Synthetic and Incremental backups to disk locations and then add a duplication template for the synthetic to write that data to tape. It is possible to do synthetics to tape - but you will need multiple tape devices so it gets a bit more complicated.

The 2 options you discuss are not correct the options actually are
1) Write full synthetic (on whatever frequency you like) - if you want a daily single tape for restore then this will have to be every night, however, how many companies keep their daily backups for extended periods - usually weekly and montly backups are kept which would be when you run your synthetic
2) Keep incrementals since last synthetic and last synthetic available (forget about the baseline as this is ONLY needed until the first synthetic is run as each synthetic becomes the beginning of the next synthetic. Note: If you need to restore from prior to last synthetic then yes you will need earlier data sets but as per point one most companies keep the weekly or monthly sets to cover this instead of keeping all the dailies.

Also the benefits of synthetic are strongly environmental - in some environments your backup window might end up needing to be longer so there will be no benefit - in others it shortens it considerably as such always recommend doing some tests against you backup windows for synthetic vs traditional options. (Full, Incremental/Differentail etc)

 

Simon_Gadsby
Level 4
Partner
So if the latest Synthetic backup is *required* in order to produce future synthetic backups, it really needs to be on disk; I get your point.

It seems unnecessarily redundant however to store two copies of the same data on disk! Assuming I only have 1 tape drive to work with, would the following work to avoid two copies of the data on disk?

1) baseline to disk
2) incrementals to disk
3) synthetic to tape
4) duplicate synthetic to disk, overwriting the baseline
--> then loop to step 2.

Whilst that might solve the 2 copies on disk issue, it still means I cannot get any sort of differentials.

Do you use synthetic backups yourself? Are you happy to wind up writing incremental backups to tape (via a duplicate job), meaning that restores could be a real pain in the butt if they have been retired from disk? What is the best way to deal with this seeming weakness other than writing full backups every night? I want to keep at least 2 weeks of backups available on tape, beyond that weeklies would be fine.

I notice however that in the 12.5 doco (p853) they say that:
"Baseline Backup - the backup template that creates the baseline backup job.
This job only needs to run once. You can run additional baseline backups
whenever you prefer, although the backup window is reduced by running a
synthetic backup instead of a full backup."
 
Would this mean that if the baseline backup runs (eg. on schedule) then it will be used as the baseline, regardless of synthetics that exist? So aren't they saying that I can choose to either use a full baseline or a synthetic as the 'seed' for future synthetic backups?

I want to try and make Synthetic backups work as I have a large amount of data which is being mounted as SAN snapshots (ie. all file data), so Synthetic backups will provide fast full backup of the vast majority of my data.

Simon.


Colin_Weaver
Moderator
Moderator
Employee Accredited Certified

If you put any part of the previous required backups on a tape then that tape must be accessible at the same time as another tape to write to.

Because you can only have one synthetic template and that template therefore can only be sent to one target device - you have a choice between writing all synthetic jobs to a B2D - or writing all synthetioc jobs to a library with more than one drive in it.

Best suggestion is

Write the basline and the synthetics into one B2D folder (and using 1 media set)
Write the Incrementals into another B2D folder (can be sthe same media set or a different one)

Make the Overwrite protection settings on the media set used by the  baseline and synthetics backups such that The baseline and synthetic backups cannot be overwritten by the next synthetic (and cannot be appended to), but can ONLY be overwritten by the synthetic after. This will result in the equivalent of 2 full backups being stored on your disk with the earliest of the two being overwritten. There is no way to get round the 2 fulls's requirement as you always have to have the previous full (which is either a baseline or a synthetic) to base the next synthetic on - so at the end of the next synthetic this will always leave you with 2 present,. BTW a duplicate job does not maintain the indexing requirements needed to run a Synthetic job based on the data in the duplicate - as such you cannot use a duplicate job as part of the sequence - it can only be used as a method of writing the same data to a different media.

The suggestion that says you can run another base line if you want, just means that you end up running a full backup to replace a synthetic backup and apart from affecting your backup window is not going to change the basic way that synthetic backups work, It just puts you back to step 1 in my original order of how the process works. So yes the syhthetic prcoess does not care whether it is a NEW baseline or a the LAST synthetic in order to run. I agree you could set this to overwrite a previous synthetic - but this would be a useless idea as you might as well just run a standard full and incremental setup and not run synthetics at all if you are going to run a Baseline every week.

Oh and I have tested synthetic backups and found that in my environment the indexing makes the incremental jobs actually take longer than a full - so no I don't run it myself any longer and hence my comments that the process is heaviliy linked to the environent being backed up and should be tested before a decision about a final solution is made.

 

Simon_Gadsby
Level 4
Partner
I have been running some tests and as you say, the synthetic backup will not succeed with a duplicate job in the sequence.

HOWEVER I have discovered that I can add a second synthetic job to the policy and use it to do the same job as a duplicate, and it succeeds. In other words, this is what appears to be working:

1) baseline to disk
2) incrementals to disk
3) first synthetic from disk to tape
4) second synthetic from tape to disk, overwriting the baseline & incrementals
--> then loop to step 2.


Are you aware of any prohibition on having 2 synthetic templates in the policy? It appears to be working fine after several tests.

So this returns me to my previous, that I do have the option of using synthetics, but I am forced to use incrementals rather than differentials...

Thoughts?
Simon.

Colin_Weaver
Moderator
Moderator
Employee Accredited Certified
No idea about running a second synthetic like you have mentioned - would say make sure you can still do a restore - and also make sure you can still catalog the tapes.

Yes you have to use Incrementals, differential jobs cannot be used - I would suspect this is more a question of why would you want to - the idea of synthetic backups is to try and shorten your backup windows - differential backups slowly get larger so you are no longer running the shortest backup. The usual reason for running a differential (instead of an incremental) is to minimise the tapes required for a restore. However,  the synthetic brings it down to one tape needed when incrementals were used  - I have heard of some users that run the daily incremental (for the synthetic)  - but then also run the synthetic daily so that they have a daily full backup but have only impacted the live network for the time it took the incremental to run.

Assuming you backup data fits on tape then the restore requirements for the different backups are:
Full Backup = 1 tape
Incremental Backup = 1 tape + the tape for each incremental since the last full
Differential = 2 tapes
Synthetic = 1 tape (if the synthetic is run regularly), 1 tape + incrementals if the synthetic is not run regularly

My guess is Symantec could see a benefit for making an incremental set into a synthetic, but could not see a huge benefit for the differential because the restore process is much simpler.

If you can provide good technical and business reasons why a Synthetic Differential would be a good idea then suggest you fill in an entry in the Ideas section (see tabs at top of forums screen) as this area is where enhancements to Symantec products or processes can be suggested and allows other Symantec Connect members to vote - which in turn provides feedback to Symantec and whether there is a lot of interest in the idea.