Start a Conversation

Unsolved

This post is more than 5 years old

3364

August 2nd, 2012 22:00

Auth. session limit

Hi all

Seeing this in daemon since last night...any idea what is this? Looks like its complaining about sessions more than 10 in that device which is set min 10 and max 60 on write device and 1 to 10 for read only device..

Auth. session limit greater than 10 for mmd# 4 on device rd=SN:DDB_dev15/_AF_readonly.

Thanks

August 2nd, 2012 22:00

Hi,

Please take a look of this link where similar issue is been reported

https://community.emc.com/thread/116265

Thanks

Shivakiran

2 Intern

 • 

326 Posts

August 5th, 2012 17:00

Hi Shivakiran

Thanks...have checked that out but different issue

23 Posts

August 29th, 2012 14:00

Hi Thierry,

I've been noticing this same error since yesterday night.

nsrd Auth. session limit greater than 32 for mmd# 1 on device dd_backleg_lsu1/_AF_readonly.

I set the target and max session to 10 and 32 on both the readonly and the rw devices.

Thierry, you ever found out what was causing this ?

Anyone else has any idea ?

Thanks!

Francois

1.4K Posts

August 29th, 2012 15:00

Restart nsr daemon. Check if the error still persists.

2 Intern

 • 

326 Posts

February 14th, 2013 13:00

nope..still same and seeing it

2 Intern

 • 

14.3K Posts

February 14th, 2013 14:00

We have noticed it during cloning when you get single session reads, but perhaps when multiple reads are going we get the same - I never paid attention to that.  I do not think we are in the same bus because we started using DD with SP4 which supports multiple reads (tested) for DD dev type devices and during those tests no issue has been seen.

23 Posts

February 14th, 2013 14:00

I'm surprised to read that multiple reads are working in SP4+. I'm running SP5 and it still won't allow more than 1 read.

If a clone is running on a DD device and then I launch a recover from the same DD, it will still say "busy" until the clone is killed off or finishes. Oh and it will also throw those "auth session limit greater than ..".

23 Posts

February 14th, 2013 14:00

^ Interesting. I have the same setup as you. I wish it worked.

As for DD Boost DFA, it only works with specific modules like NMDA or NMO, from what I understand. In my case, all the oracle RMAN backups go directly to the DD instead of going through the storage node using DFA. Everything else (standard FS backups for example) needs to go through the storage node.

Networker 8.0 adds support for simple NW clients (filesystem) to dedup directly to DD:

From NW 8.0 release notes:

The EMC NetWorker 8.0 releases provide significant functional enhancements, including

support for the following:

  • Direct File Access (DFA) enhancements, including support for DD Boost and standard AFTD devices, and the ability to bypass the NetWorker storage node and send data directly to a Data Domain system.

From the data domain integration guide (7.6), see table 2 on page 17. It states that DFA backup is not supported for filesystem backups in earlier version than 8.0.

UPDATE: I just re-read the table and your post.. it indicates that recovers support DFA. Oops. I will try multiple recovers at the same time then, I only tried with clones running, and that fails. Maybe if there's no clones, multiple recovers would work..

23 Posts

February 14th, 2013 14:00

For me, it happens every time more than one recover and/or clone session is trying to read from an AFTD device, which is a limitation of NW 7.x (only one recover/clone session on AFTP at a time).

2 Intern

 • 

326 Posts

February 14th, 2013 14:00

LOL...but nope

2 Intern

 • 

14.3K Posts

February 14th, 2013 14:00

I can bet half of my ass that you have nsrmmd core dumps when that happens

2 Intern

 • 

14.3K Posts

February 14th, 2013 14:00

lapfrank1234 wrote:

I'm surprised to read that multiple reads are working in SP4+. I'm running SP5 and it still won't allow more than 1 read.

If a clone is running on a DD device and then I launch a recover from the same DD, it will still say "busy" until the clone is killed off or finishes. Oh and it will also throw those "auth session limit greater than ..".

That's not what we see.  We use Linux media servers and all reads are done from them (cloning and restores). Further, since we use DD Boost DFA, streams are read from DD directly so this might be part which enables us to use multiple reads. One of the reasons why we tested this was also because during nsrim check, we would see locks on device and no restore would be able to run - this is also fixed since SP4 (again, at least when using DD type of devices).

2 Intern

 • 

326 Posts

February 14th, 2013 15:00

Slight distraction from discussion

If DFA works on specific module only...how ndmp_dsa can be configured to use DFA (excerpt from 7.6.3 admin guide)

Data write technologies

The nsrdsa_save program, on execution, establishes a save session with the

NetWorker server and listens on a port number. After a DSA receives data from the

NDMP client, there are three different technologies that a DSA can use to write the

data to a NetWorker storage device.

◆ Immediate save technology

The DSA reads the data by using TCP/IP into shared memory. The nsrmmd

program writes the data to a storage device from shared memory. Virtually all the

data from the NDMP client goes directly to nsrmmd.

◆ Direct file access (DFA) technology for a backup to disk

DFA directly writes the data directly to disk, bypassing the storage node

(nsrmmd program). The storage node only plays a role in loading the volume.

This is a highly efficient and high performance approach.

◆ Nonimmediate save technology (or remote save)

The DSA reads the data by using TCP/IP into a local buffer and transmits it to the

nsrmmd program that is also using TCP/IP. The nsrmmd program writes it to the

storage device. This approach is inefficient and has slow performance.

2 Intern

 • 

14.3K Posts

February 14th, 2013 15:00

Yes, DFA in NW7 can be used only with specific module.  These are vast majority of my backups (and restores). I believe storage node itself is using DFA connection to DD and in that way is allowing multiple reads to kick in in SP4 and SP5.  The way we did a test was rather simple: we did two big backups and then run restore in loop for both - and both would run ok all the time - no issues.  Indeed, nw8 adds DFA to client backups, but I don't need that at the moment as databases and logs are 80% of the backup volume I have.  Going back to multiple reads, if you have multiple requests for reads, then media server itself will use DFA to pull back data and from there you get two indepedent streams to the different clients - my guess is, this is how they did it.

2 Intern

 • 

14.3K Posts

February 14th, 2013 15:00

I don't think DD's DFA can be applied to any NDMP backup (unless perhaps VNX if they have DDBoost library, but to my knowledge this ain't officially there yet). So, NDMP backup will always go through nsrmmd unless it knows how to communicated with Boost device and that's possible only if local ndmpd will have a clue of that API (of course, you could always do a trick with 3rd party ndmpd server which runs on open system server and which has ddboost library, but I would not call that direct save). Now, around EMC World time things get announced and timing seems to be right for NW 8.1 (which kicks with official beta next month) so whether this will be in at the end or not - who knows. But we may already learn about this somewhere in May.

No Events found!

Top