NBU Accelerator for DataDomain

[Masami_Katsuno]

Accelerator is now be supported by DataDomain with Plig-in v2.6 on HCL.

I would like to confirm that physical full backup will not have to be done forever at this accelerator with DataDomain. 

In this case, does NBU detect the changed block on the client and send them to DataDomain and then do opt-synthetic Full?

Regards,

Masami Katsuno

[Nicolai]

Nobody knows how EMC pulled off the accelerator support for Data Domain.

There is a manual out there how to configure it, but it does not explain how it works. I had in mind to open a support ticket with EMC to find a answer.

Best Regards

Nicolai

[Marianne]

It would be nice if someone could test this and let us know....

[trv]

Accelerator is working with DDOS 5.4 and DDBOOST plugin 2.6, you will have to ask DataDomain support to grant you access to restricted access versions of both I am afraid. Anyway, we are testing datadomain + nbu accelerator with NetBackup 7.5.0.6, ddos 5.4.0.7 and ddboost plugin 2.6.1.0 and it is working just fine. You can skip 'get new mappings file' part too it seems, it is working with stock version ... in 7.5.0.6 at least.

[Nicolai]

Can you explain how it work in detail (not how to configure it)  ?

[trv]

It works as any other device with accelerator option. Nbu accelerator is just synthetic backup (possible since ddos 5.2) with some sugar on top from DataDomain point of view anyway, so there is nothing DataDomain specific as far as I know. Guess you will have to check generic Accelerator documentation.

[Marianne]

What we really would like to know - does the Client track log work so that only changed blocks are sent to media server? Or changed files? Or all of the data?

[trv]

Well I can't tell exactly, but it seems to work like with MSDP accelerator enabled backup without clientside deduplication. Which means bpbkar is sending changed blocks only (nothing OSTish on the client needed) and media server + ddboost plugin is doing the magic. ... 10/29/2013 08:00:05 - Info bpbrm (pid=53735) accelerator enabled ... 10/29/2013 08:00:15 - Info bpbkar (pid=53740) accelerator sent 39874560 bytes out of 80112128 bytes to server, optimization 50.2% ...

[Yasuhisa_Ishika]

Process overview of Accelerator backup is documented in NetBackup 7.5 Administrator's Guide(page 566 for Windows and page 646 for UNIX). Media server interact with storage server through OST plugins, and operations in this part is similar to optimized synthetic backups. Nothing mention to variety of client operations here, so i guess amount and type of data sent from clients to media servers is not depend on OST plugin type - only changed block are sent as documented.

[Masami_Katsuno]

Should I install the Plug-In v2.6 on Storage Server(Media Server) or Accelerator client or both?

[Yasuhisa_Ishika]

I bet on media server. storage server is DD in this case.

[Marianne]

As per actual results seen in this post, it seems to me that more than just changed blocks are sent. 50% optimization seems rather low, don't you think?

[Yasuhisa_Ishika]

I'm not sure if 50% optimization is lower than expected in trv's use case. bpbkar just says 49.77% of data sent to server.

[trv]

Ddboost plugin should be on media server as always. Installing OST plugin on client is not supported yet and I am not sure it ever will be (which is kinda sad ... no client side dedupe unless you are using puredisk or willing to convert client to media server). Lines from detailed job log I posted were from incremental backup, here are some numbers from full backup: ... 10/30/2013 08:00:05 - Info bpbrm (pid=64776) reading file list from client 10/30/2013 08:00:05 - Info bpbrm (pid=64776) accelerator enabled 10/30/2013 08:00:05 - started process bpbrm (pid=64776) 10/30/2013 08:00:06 - Info bpbrm (pid=64776) starting bpbkar on client 10/30/2013 08:00:06 - Info bpbkar (pid=64781) Backup started ... 10/30/2013 08:00:17 - Info bpbkar (pid=64781) accelerator sent 77404160 bytes out of 8450501632 bytes to server, optimization 99.1% 10/30/2013 08:00:17 - Info bpbkar (pid=64781) bpbkar waited 0 times for empty buffer, delayed 0 times 10/30/2013 08:00:18 - Info bptm (pid=64782) waited for full buffer 1 times, delayed 26 times 10/30/2013 08:01:19 - Info bptm (pid=64782) EXITING with status 0 <---------- ... Accelerator effectiveness could vary indeed, but I am quite sure it doesn't depend on target storage (DataDomain in my case).

[Nicolai]

I opened a support case at EMC to get a official answer how accelerator works with Data Domain. Level 1 support could not provide an answer. Escalation needed :(