This post is more than 5 years old
3 Posts
0
3130
February 28th, 2013 11:00
NetWorker 8.1 and Data Domain backup policy
Hello,
Please let me know what would be the best backup policy for this environment:
1 NetWorker server
2 Data Domain devices: 1 in the same site as the server, 1 at the remote site for DR
60 Windows servers
40 Linux servers
I was thinking to use:
1 Group for BMR
1 Group for Windows
1 Group for Linux
Each group will backup to a different media pool containing one Data Domain device. (In terms of performance, would it make sense to have more that 1 Data Domain device in each pool ?)
The backup level would be Full on Sunday and Incremental (during the evenings) on the other days of the week. Do you think it would be a good idea to use the synthetic full in the middle of the week?
The data would be cloned during the day to the remote site.
Would Client Direct backup make sense in terms of performance? Should it be used for servers with a lot of files?
Thank you!


ble1
6 Operator
•
14.4K Posts
•
56.2K Points
0
March 1st, 2013 02:00
Let me start from the end - this is a good question - and something I'm actually trying to get clear answer myself at ATE session too (click here). I have another question semi-related in respect to max nsrmmd count, but that one is still not visible as it is pending moderator approval (doh). Anyway, when using direct writes to DD that probably would not matter (let's assume that, I can't say if that is the case or not, but if each device is file within same SU than you depend on performance management of that file). NW documentation is somewhat confusing on this as different places tend to suggest different approaches, but overall I would always go for several devices - in worst cae scenario I will to make sure that if something happens to one device, others are not involved (DD and NW wise). Further, NW docs seem to suggest best practice to be up to 10 sessions on DD device. You have 100 clients. Assume you make groups with 10 clients, starting at 8pm first, second at 9pm, etc... even so one group would have more than 10 sessions (more likely 60 sessions at least). Of course, you can limit that, but why limit anything when you can simply spread it.
In my own case, I do not differentiate between clients who have X amount of data or Y amount of files. What I do is following:
- each group will have 10 clients
- each client will have full on working day (n+1 compared to previous one)
- fs backups do not run during weekend
So, it looks as this, for Group_A
name: Client_1 Schedule: Full_Monday_skip_weekend_rest_Incr
name: Client_2 Schedule: Full_Tuesday_skip_weekend_rest_Incr
name: Client_3 Schedule: Full_Wednesday_skip_weekend_rest_Incr
name: Client_4 Schedule: Full_Thursday_skip_weekend_rest_Incr
name: Client_5 Schedule: Full_Friday_skip_weekend_rest_Incr
name: Client_6 Schedule: Full_Monday_skip_weekend_rest_Incr
name: Client_7 Schedule: Full_Tuesday_skip_weekend_rest_Incr
name: Client_8 Schedule: Full_Wednesday_skip_weekend_rest_Incr
name: Client_9 Schedule: Full_Thursday_skip_weekend_rest_Incr
name: Client_10 Schedule: Full_Friday_skip_weekend_rest_Incr
That way, I never have more than two clients doing full backup withing group run. And most importantly my weekends are free so no one has to keep an eye on backups. Well kind of - I run database archive logs then and full for clients which are backing up over WAN. But overall, load is spread and works perfectly (in my own case). You plan to focus highest load on weekend which is something I would never do.
ble1
6 Operator
•
14.4K Posts
•
56.2K Points
1
February 28th, 2013 12:00
There is no need to separate clients in groups per their OSes. Actually, you will find that mosly Linux boxes are faster in completing than Windows boxes (less OS data). I would groups them as per data volume and tried to spread the load across backup window. You will need more than 2 devices for sure. Client direct makes sense in terms of performance (and also load to storage node). I don't think you need synthetic full with weekly fulls in place (unless big data change in incrementals and slow link between server and client).
dan_apostu
3 Posts
0
March 1st, 2013 02:00
Hi Hrvoje,
Would you group the clients according to the amount of data (small amounts with big amounts) or by number of files?
Why do you think that more than 1 device per pool will be needed? I'm asking this because all the information that I received from Data Domain is revealing that from their side the number of devices is not related to performance.
Will NetWorker handle better the load when addressing multiple devices? Is it a relation between the number of nsrmmd processes and devices?
Thank you!
dan_apostu
3 Posts
0
March 13th, 2013 04:00
Hi Hrvoje,
This is great information!
Only one thing, don't you have clients that span a full backup multiple days because of the size of the data? In this case files that are change since the backup begun are not protected. I wanted to do the full backups in the weekend so that I minimize the amount of files that are changed and not covered by the backup.
ble1
6 Operator
•
14.4K Posts
•
56.2K Points
0
March 13th, 2013 08:00
You will always have files changed after backup has started or finished - no escape there. This is why most of the important backups (production) run off the business hours. If you work hard and get overtime by working after business hours, no backup for you