Unsolved
This post is more than 5 years old
5 Posts
0
2140
Replication/duplication initial starting
Currently using Netbackup 7.6.0.2 with Data Domain units for replication to remote central storage.
Know it is best to seed the target unit because it allows better replication times. If seeding is not possible which way is better.?
1.Using SLP create a full backup job for backup and replication and let this job completely finish before starting another job. This may take a while.
2. Using SLP create the backup job and allow multiple weekly jobs to run.
Tried the second method and ended up with lots of jobs running. Seemed to me that the many jobs were interfering with each other.
One of my group cancelled the earlier jobs. This seems to be counterproductive as the data transferred is probably not usable.
My understanding of the way DD Boost works with the SLP replication is that the source machine works with the target machine to send only new data. Since this is the original set of data this will take a while. Is there a better way to do the seeding across the network? Would just doing a backup to the target device be better than using SLP.
We actually do a monthly (long term storage) and weekly-full (short term storage) and am considering only doing the monthly backup because of the time it is taking to complete the replication/duplication process.
tks
ed
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
0
September 29th, 2014 08:00
Ed,
I have a few questions for you to determine this can work for you?
1. What model DDRs are you using? Based on the model we can determine how our options based on system resources or stream counts?
2. Which version of DDOS are you running?
3. How fast is the WAN between DDRs? Based on the network speed, we may have limiting factors for replication throughput.
4. Do you have firewalls, QoS, WAN accelerators, Packet Accelerators in the network path between the DDRs? This information will determine if you are achieving the expect max throughput. These devices can actually slow down DD Replication.
Based on the Symantec Version [7.6] you are using, are you using the standard SLP configuration or NBU AIR?
I haven't had a chance to compare the 2 options for performance, but I thought I should ask.
References:
https://support.emc.com/docu50024_Data_Domain_Boost_for_OpenStorage_2.6_Administration_Guide_.pdf?language=en_US - EMC DDboost Administration Guide (2.6 Plugin)
https://support.emc.com/kb/182214 - NBU AIR Troubleshooting Guide
https://support.emc.com/docu48802_Symantec_NetBackup_OpenStorage_Best_Practices_Guide.pdf?language=en_US - NBU - OpenStorage Best Practice Guide
ed_smith
5 Posts
0
September 29th, 2014 14:00
1. Target is DD640 Origin DD2500, DD640, DD160
2 5.4.2.1-423209
3 depends on location 512 to 5 mb Limited during working hours. Varies from location
4 all over the place
standard slp
Think am making this too complicated. Know that prestaging data is the best way to go. However; in places where this is not possible should I let the first full replication (each server) job finish before starting more? All other questions follow from that.
tks
ed
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
0
September 30th, 2014 05:00
I agree, you may be making this more complicated than it needs to be.
First suggestion: (engage support)
I would make would be to open a service request with Data Domain. Support will be able to look at detailed information on the systems, auto supports and within NBU to provide you with a suitable strategy.
Second suggestion: (Environment Tuning)
1. Netbackup/DD Clone Best Practices - Is the configuration optimized. Number of devices, Correct Streams counts
2. Network Configuration - Is the network causing potential problems?
3. DDR resources - Are we efficiently using the Data Domain Resources?
What SLA do you have on your data? Do you have separate retention policies for source and destination DDRs? Is this separate for different types of backups(data types)? It may be possible to adjust the DDR devices settings to process more clones concurrently, but this need to be monitored to make sure you are not running into problems due to being out of resources (Network or DDR).
Commands to be aware of:
ddboost show connections - displays the # of concurrent connection currently running
ddboost file-replication show performance interval 5 - show file-replication performance at an interval of 5 sec. Good to use to see the speed of replication.
net iperf server
net iperf client
traceroute
ping
When you have varied bandwidth, you must monitor the replication traffic. The network could cause RPC timeouts which forces data to be resent or force NBU to perform retry the clone job. This gives the appearance that replication is slow. This can only be verified by review the logs. (Support can assist with this)
Tools to help with bandwidth verification:
ping
traceroute
iperf - https://support.emc.com/kb/180574
Once the network bandwidth is understood. You will need to be aware of the replication stream counts. The DDRs models have different numbers of allocated replication streams based on the resource in the system. You always want to try to match or have the system with most replication streams at the destination or you could run out of streams (resources)
https://support.emc.com/kb/180681 - Stream Counts FAQ
https://support.emc.com/kb/184683 - Support Stream count per DDOS 5.4