ONTAP Discussions

NDMP Backup 2MB/min

oweinmann
20,528 Views

Hi,

is there a limitation (file count) for NDMP? We see a very very poor Performance with NDMP Backups of a 2040c with BackupExec 2010r3. We have not evaluated a different backup software yet so we are not sure if this is the bottle neck.

The volume we are trying to backup consists of several qtree's and has a total around 16 Mio Files.

Regards,

Oliver

1 ACCEPTED SOLUTION

oweinmann
14,158 Views

Hi,

we finally have backup jobs running with backup exec at a very good average speed of 3.8GB/min. So we need 13 hours for around 3.1TB. We have two tape drives so we could even split the jobs.

The key was to restructure the folder structure. Before we had just one large volume Projects with up to 14 million files. I have now started to break it down in ProjectsAC to ProjectsXZ and used snapmirror to move the qtrees. This had a signifficant impact on the backup speed.

Problem solved!

View solution in original post

25 REPLIES 25

andrc
18,872 Views

I'm assuming the filer is directly connected to the tape device and data flow isn't passing through a media server?

As a start you can issue `sysstat -u 1` from the CLI and check for any obvious bottlenecks i.e. CPU, disk utilisation plus you'll get a clear view of approximate KB/s the filer is writing to tape.

aborzenkov
18,872 Views

The first step dump does is to build full file list; with large number of files it can be quite time consuming. You can test the theoretical dump speed by calling it directly and dumping to NULL:

https://kb.netapp.com/support/index?page=content&id=1011894

Watch how much time each of steps takes. You should also be able to see dump logs in /etc/log/backup.

oweinmann
18,872 Views

Hi and thanks for the quick responses:

I'm assuming the filer is directly connected to the tape device and data flow isn't passing through a media server?

- Yes it is directly attached to our FAS2040 via FC.

I have now snapmirrored the largest qtree (1.8mio files) to a new volume and started backup exec. Let's see how long it takes. If this is already an issue we have to review the space usage.

Regards,

oweinmann
18,872 Views

sysstat shows a very high disk usage (80-97%) when running the backup.

oweinmann
18,872 Views

Ok, once the backup has really started the disk usage drops below 30%. It is just sitting around for about 15 minutes. I guess this is when the list of files to backup is created.

jwrobertson
19,447 Views

We've had problems similar to this when we back up volumes that contain large numbers of files.  It has to crawl the file list before it actually starts backing up, and sometimes that can take hours before it actually starts moving any data.  Commvault supposedly now has a solution that allows it to skip that crawl by using snapshots and only looking at changed data.  Another option that NetApp presented to us is doing a volume based backup.  That avoids the file crawl, but doesn't allow individual file restores... you have to restore the entire volume.  Might be an option if you can use snapshots for individual restores and the volume backup for major disasters.

jwrobertson
19,447 Views

I should also add that this is not a problem that is unique to NetApp.  We had the same issues doing NDMP backups from a Celerra as well.

oweinmann
19,448 Views

Hi,

thanks for the recommendations. Is there a trial version of commvault available? I had recommended commvault in the past to our boss but we had backup exec for several years now and it is quite ok. I will now evacuate all qtrees wth more than 600.000 files to a new volume and create a separate backup job for it. We have two LTO 5 FC Tape drives, so we can split it up.

eric_barlier
19,448 Views

hi,

You said: "The volume we are trying to backup consists of several qtree's and has a total around 16 Mio Files."

Thats your problem right there. NDMP is not good for lots of small files. Off the top of my head I think NDMP backups

have to 5 phases (or was it 6) and the first 3 ones are just for scanning the file system. Imagine having to scan

that many files on a very small controller. You havent got the CPU to deal with this I reckon, even if you did it would

take ages to back this up.

The way to work around this is to use snap products as backup methology. As mentioned commvault integrates nicely

with netapp snaps.

Cheers,

Eric

oweinmann
16,014 Views

I will try to contact commvault if  they can supply us with a trial version to test the ndmp backup.

glen_eustace
16,015 Views

We are using EMC Networker at are seeing extremely slow thoughput as well.  We get 7Mb/s regardless of whether the volume has many files or one large file of a similar size.  A CIFS mount to the same volume and we get 70Mb/s.  There is something extremely sucky about NDMP.  To date we have had no suggestions from either EMC or IBM (N-Eries) that has made any difference at all.

aborzenkov
16,015 Views

I had several installations with NetWorker and backup speed was OK. Restore was different matter though ☹

Incidentally, all installations were using 3-way backup (via NetWorker DSA) and not direct backup to tape. I constantly observed full wire speed (120MB/s for 1Gb/s interface). But these all were database installations with relatively small number of large files, 10-20GB each.

glen_eustace
12,976 Views

This is our scenario as well.  We have a DLT4 jukebox on a storage node.  Throughput is abysmal when using NDMP.  As mentioned in my previous post, using CIFS as the transport, with all other hardware the same we can max out the network connection, so we have ruled out the tape and networker.  Even trying to multiplex a couple of NDMP backups the trhoughput is unacceptably slow it seems that the filer is simply incapable of pushing the data out any faster.

This filer is at our DR site and has no customer access.  It had been our intent to offline the snap-mirror detination volumes using NDMP to tape.  We should be able to get the entire IO capacity of the filer.  We have almost given up and are currently trying to find alternatives

aborzenkov
16,014 Views

Oh, and the first thing that comes to mind is to check fragmentation level ☺

oweinmann
16,014 Views

Hi,

we will now evaluate commvault to see if it works. Hopefully we get the trial version very soon.

Cheers,

Oli

radek_kubka
16,014 Views

Hi Oli,

Make sure you test SnapMirror to Tape (SMT), which is supported by CommVault - that way the whole issue of crawling file system structure goes away (it is like VSM basically).

Regards,

Radek

aborzenkov
12,977 Views

Does it support per-file recovery?

radek_kubka
12,977 Views

Good point.

I don't think it does, as this is basically a full NetApp volume copy on a non-NetApp tape.

oweinmann
12,977 Views

No it doesn't support it. We had a discussion about Backup with our Netapp Reseller regarding backup and they recommended this for a disaster recovery where you want to restore a whole volume and not individual files.

oweinmann
13,102 Views

Ok, finally we have now received our copy of commvault simpana 9.0 64Bit. I'm currently testing ndmp backup performance. The volume with the millions of files still doesn't backup faster. I started the backup about an hour ago and it is still sitting there with 0% in progress. I read about their snap protect feature that should solve the issue with the million files. I have enabled it but the backup is again sitting there since 20 minutes with 0%.

Public