Unsolved
1 Rookie
•
7 Posts
0
276
October 17th, 2022 06:00
clone 4 streams concurrent from DD to one tape volume
Hi,
I have a issue:
NetWorker 19.6.1 on RH...
I set the tape target-sessions to 4, and max-sessions to 6 on a LTO-7 tape device.
Backup- and Clone-pool parallellisme is "infinity" (0).
I have 4 backup-pools containing the data I want to make a clone from.
I added 4 DD-devices with RO flag to handle any type of reading the backup-data for each backup-pool (so 16 RO devices).
I did not change the access weight of the RW and RO devices (all are 1).
I have a batch to build 4 lists of savesets containing balanced in size a unique list of savesets containg a mixed list of savesets of the data from the 4 pools.
It is planned to initiate the nsrclone process spawned for each of the 4 lists to initiate the concurrent cloning of 4 streams to a single tape-volume.
=+=+=+=
If either RPS is enabled or disabled and I initiate the cloning via the GUI, I have only 1 streams @ 60 MB/s (data dependend).
if RPS is disabled and I spawn 4 nsrclone jobs via the CLI, I can see only one session running at about 60MB/s (data dependend).
If I enable RPS (clear the not disabled flag) and spawn 3 or 4 nsrclone jobs, I can see up to 4 streams and running @ 240 .. 500 MB/s (data dependend)
Due to the amount of data, cloning @ 240 MB/s will still take 6 working days.
Unfortunally, as far as I can recall RPS was ment for DD-2-DD replication and not DD-2-Tape ...
Did I miss a update regarding the RPS-feature?
I hope the solution is supported...
bingo.1
2.4K Posts
1
October 17th, 2022 14:00
Hi Gert,
according to IBM's specification, the native device speed for a LTO7 is about 300MB/s. IMHO achieving about 80% of this value in a true environment is not that bad. If you really want to verify the max. performance, may I suggest that you test the speed with a NetWorker 'bigasm' backup. But of course this does not represent the production environment.
Whether you use RPS or not can become tricky - if you clone from DD to tape. I remember that we needed some time to finally decide not to use it (to check the option Disable RPS). However, I found out some days ago that it will reduce the recover time from a DD cloud tier to the active tier using an ECS as clone device. In my case, it reduced the recall of a save set from about 55 mins to 15 secs as only metadata had to be transferred.This supports your assumption - but it does not help in your case.
Back to your scenario. The only possibility I see is to add more tape drives and to run multiple clone processes simultaneously. We had a similar issue but we were lucky to have 14 tape drives. The solution was pretty much the same:
- we created save set lists for each DD volume.
- we started one nsrclone command for each save set list with a delay of 5 mins in between.
This worked fine for years.
Hope this helps.