Savegroup Parallelism

Anuj, think of the drives as a one large device........
drive 1 has 4 parallelism
drive 2 has 4 parallelism

so when your backups start, the 1st FOUR backup streams will take the first 4 available sessions, which will be drive 1 since it is set at 4........

Now if you set both Drives to a parallelism of 1, then then your 1st backup starts, it will start as
Drive 1 = 1st stream
Drive 2 = 2nd stream
Drive 1 = 3rd stream
Drive 2 = 4th stream......

it will continue doing this as the drives are at "1"......set it to 2, and you get the 1st 2 on the 1st device, the next 2 on the 2nd device, and then the next 2 are on the 1st device etc etc....

if your 2nd group is also set at 4, then it it probably only taking 2 drives, due to the 1st group still having a session or 2 running at the same time.......if there are no sessions running at all when the 2nd group starts, its unusual that it will use both drives concurrently if they are set to 4.....

To add: target sessions are not fixed number of sessions going to device, but rather load balancing one. Max session numbers are fixed value which says max number of sessions allowed going onto device.

I believe when you set group parallelism to 4, it should generate only 4 streams - the first four it receives form the clients it has in its configuration.If target session for the drives is 4, one of the drives would start writting these 4 streams.

In case target session was set to 2 each, it would spread the 4 group streams to 2 each for the two drives. Similarly, if GP is 8 & TS is 4, it should distribute 4 from each group to the 2 drives.

It is very strange that GP is 4 and TS is 4 but still is picks up 8 streams and uses both drives.

Yes, you should never see more than 8 streams at the time. However they might use different tapes as I believe nsrmmd allocation happens at early stage even those streams are queued up. However usage of two drives should be seen at later stage and not immediatelly. WHat version of Nw is that? I remember some versions had bug with savegrp parallelism (for some reason I originally misread savegrp parallelism in this thread for server parallelism so I didn't pay attention to what the problem is really).

You lost me now... you said initially that second group takes 2 drives...

Infact the other group works fine, takes only 4 streams even if the two drives are free and sufficient media is available.

We got this issue with 7.2.2 earlier. Currently we are using 7.3 Jumbo 8 and plan to upgrade to 7.3.2 by Tuesday.

Sorry, the first one works fine and the second one creates a problem.

I was just wondering would it matter if we have 5-6 clients each with 3-4 save sets. The group that works fine has only one client with 4 savesets. Normally, savegroup parallelism should take care of this.

Number of clients should not have any impact as savegrp parallelism applies to whole savegrp (so number of clients should be irrelevant). I guess both groups use same pool, right? In such case sounds as something is broken and involving support further would be recommended.

Incidentally, they are using different Media Pools.

I will contact support on Monday and let us see how it goes.

Since you are using Pools why don't you limit Pool A to Device 1 and Pool B to Device 2?

Edit your Pool settings and check the box for the device. Then the Pool can only mount tapes labled for its Pool on the device you select.

I don't appreciate the idea of binding devices to pools or groups. You need to be sure that the drive is working and this group finishes off well before the third group otherwise if they overlap, the third group waits for long.

I don't have this issue because I have Libraries with dual drives. I use one drive and Pool for File Level Backups and one drive and Pool for DB backups. Separating DB backups from non DB backups is a Networker best practice. When I label tapes I have a 60/40 split of DB to File backup usage.

File backups go Pool A, Drive 1 and DB backups go to Pool B drive 2. If I ever have a need to mount a Tape from one pool on both drives then I temporarily check the box to allow the tape from the respective pool to be mounted on both drives.

I backup TB's of data on a daily basis I this setup has worked flawlessly for several years.

You must be having two groups only - one on each drive but we have multiples here. We do have DBs in a different pool then file system but need to know why the standard parallelism system not work.

Hi Anuj,

Why are you running two groups at the same time using two separate media pools?

Run one group with a parallelism of 0 (Default value meaning no restriction on the number of streams.) and use one media pool which has both devices selected. Set the parallelism on each device to 2. Also set each client's parallelism value equal to its total number of savesets and set the server parallelism equal to 32 (This is the maximum number of streams for a stand-alone Networker server. Clusters containing a Networker server and multiple storage nodes can have a maximum server parallelism of 512.) If your devices can't handle at least 10MB/sec transfer speeds each, set the server paralism to smaller and smaller even numbers (30,28,26,24,22,20,18,16,14,12,etc) until you can maximize the performance with the highest number of saveset streams. The two devices will have an equal number of streams until the number of incomplete streams drops below the server parallelism value.

If you have a tape library with two devices, make sure you set the library parallelism value to 2 so that simultaneous operations can be performed on them.

I want more servers to send data for backups at the same time but on different media.