Heino,
When the backups are running have you confirmed its going over the 10G
interfaces (end to end -- do a packet trace)? I had a heck of a time
getting it working without intercluster interfaces.
Get harvest/OCUM running and check if there are bottlenecks when the
backups kick in. Also look up multi streams.
I had a similar issue to what you are describing -- it was defaulting to 1G
mgmt, I had to configure a cluster wide 10G management interface and define
it before it started working -- the support recommended node mgmt ints did
not work.
Regards,
Douglas
On Wed, Jun 2, 2021 at 10:26 AM Heino Walther <hw@beardmann.dk> wrote:
> So maybe a volume reallocate would make it speed up?
>
> One of our systems is quite new and has not been over 80% on its
> aggregates, so I would think that reallocate would not make much sense…
> also it is a snapvault target and no snapshots have been deleted yet, so
> data has just been added over time…
>
>
>
> I would guess the issue can be that if the volume is dedupe enabled, the
> data will no doubt be “spread” around the disks somehow…. And these volumes
> have about 70% dedupe rates ????
>
>
>
> I have read a bit about snapmirror to tape, maybe that is the approach?
> Has anyone tried this?
>
>
>
> I do not have much faith to reallocation… so we are looking into FC
> attached if it is possible via FCoE (which I doubt)?
>
> Another approach would be to mount up the volumes (they are NFS) to a
> Linux hosts that has a NetBackup Client installed, and then backup the data
> from there…
>
> We have so much data, that we would need to do a full, and then only do
> incrementals from there… (the volumes are vmware datastores)
>
>
>
> (I know that other setups are possible, like FabricPool etc.. and that
> might be the way forward in the future)
>
>
>
> /Heino
>
>
>
>
>
>
>
> *Fra: *Rhorer, Kyle L. (JSC-OD)[THE BOEING COMPANY] <
> kyle.l.rhorer@nasa.gov>
> *Dato: *onsdag, 2. juni 2021 kl. 16.07
> *Til: *Heino Walther <hw@beardmann.dk>, toasters@teaparty.net <
> toasters@teaparty.net>
> *Emne: *Re: [EXTERNAL] NDMP Backup speeds
>
> We have a scenario very similar to yours — NDMP over 10GbE with NetBackup.
> I was seeing backup speeds from our FAS2650 of several hundred MB/s in the
> beginning (first several months) but that has gradually slowed over time to
> about 35 MB/s. It is surely disappointing. It’s also disappointing that the
> Filer doesn’t have enough interfaces for us to attach the tape drives
> directly like we were doing with our old FAS3040 but if we could get back
> to the several hundred MB/s speeds that we had in the beginning then it
> would be an overall win.
>
>
>
> *From: *Toasters <toasters-bounces@teaparty.net> on behalf of Heino
> Walther <hw@beardmann.dk>
> *Date: *Wednesday, June 2, 2021 at 08:46
> *To: *"toasters@teaparty.net" <toasters@teaparty.net>
> *Subject: *[EXTERNAL] NDMP Backup speeds
>
>
>
> Hi there
>
>
>
> We are doing NDMP backups from out FAS2750 with NL-SAS disks onto LTO8
> tape drives via NetBackup.
>
> For other reasons we are doing this via 10G Ethernet, and not directly
> connected tapedrives to the NetApp controllers.
>
> We are seeing backup speeds of about 40-50MB/sec. which is kinda slow, the
> backup is a Windows host that is doing nothing… no CPU or memory load as we
> backup.
>
> The LTO8 drives should be able to do much better (350MB/sec. before
> compression)…
>
> The aggregates are RAID-TEC with 17 drives in it. We are on ONTAP 9.7.
> We are not seeing a lot of load as we are running the backups…
>
> Could it be that the protocol itself it as cause here?
>
> In the “old days” we did NDMP backup directly from the controllers to the
> tape drives.
>
> We are now thinking of converting two 10G ports to fibrechannel ports… if
> possible…
>
> So I have two questions ????
>
>
>
> The FAS2750 is configured with RJ45 10G ports, but would it be possible to
> use FCoE and somehow connect this to a FC/Ethernet switch and be able to
> connect it to the tape drive which is 8G FC (LC Connectors) ?
>
> (I have not worked that much with FCoE… if the FAS2750 was configured with
> CNA-Ports, it would have been easy to convert that into FC)
>
>
>
> Has anyone tried to backup NL-SAS volumes similar to this over directly
> connected NDMP, and what kind of speeds are you seeing?
>
>
>
> Finally if anyone have any idea if the NetBackup setup can be tweaked
> somehow, that would also be useful… (we have tweaked it a bit)
>
>
>
> /Heino
> _______________________________________________
> Toasters mailing list
> Toasters@teaparty.net
> https://www.teaparty.net/mailman/listinfo/toasters
When the backups are running have you confirmed its going over the 10G
interfaces (end to end -- do a packet trace)? I had a heck of a time
getting it working without intercluster interfaces.
Get harvest/OCUM running and check if there are bottlenecks when the
backups kick in. Also look up multi streams.
I had a similar issue to what you are describing -- it was defaulting to 1G
mgmt, I had to configure a cluster wide 10G management interface and define
it before it started working -- the support recommended node mgmt ints did
not work.
Regards,
Douglas
On Wed, Jun 2, 2021 at 10:26 AM Heino Walther <hw@beardmann.dk> wrote:
> So maybe a volume reallocate would make it speed up?
>
> One of our systems is quite new and has not been over 80% on its
> aggregates, so I would think that reallocate would not make much sense…
> also it is a snapvault target and no snapshots have been deleted yet, so
> data has just been added over time…
>
>
>
> I would guess the issue can be that if the volume is dedupe enabled, the
> data will no doubt be “spread” around the disks somehow…. And these volumes
> have about 70% dedupe rates ????
>
>
>
> I have read a bit about snapmirror to tape, maybe that is the approach?
> Has anyone tried this?
>
>
>
> I do not have much faith to reallocation… so we are looking into FC
> attached if it is possible via FCoE (which I doubt)?
>
> Another approach would be to mount up the volumes (they are NFS) to a
> Linux hosts that has a NetBackup Client installed, and then backup the data
> from there…
>
> We have so much data, that we would need to do a full, and then only do
> incrementals from there… (the volumes are vmware datastores)
>
>
>
> (I know that other setups are possible, like FabricPool etc.. and that
> might be the way forward in the future)
>
>
>
> /Heino
>
>
>
>
>
>
>
> *Fra: *Rhorer, Kyle L. (JSC-OD)[THE BOEING COMPANY] <
> kyle.l.rhorer@nasa.gov>
> *Dato: *onsdag, 2. juni 2021 kl. 16.07
> *Til: *Heino Walther <hw@beardmann.dk>, toasters@teaparty.net <
> toasters@teaparty.net>
> *Emne: *Re: [EXTERNAL] NDMP Backup speeds
>
> We have a scenario very similar to yours — NDMP over 10GbE with NetBackup.
> I was seeing backup speeds from our FAS2650 of several hundred MB/s in the
> beginning (first several months) but that has gradually slowed over time to
> about 35 MB/s. It is surely disappointing. It’s also disappointing that the
> Filer doesn’t have enough interfaces for us to attach the tape drives
> directly like we were doing with our old FAS3040 but if we could get back
> to the several hundred MB/s speeds that we had in the beginning then it
> would be an overall win.
>
>
>
> *From: *Toasters <toasters-bounces@teaparty.net> on behalf of Heino
> Walther <hw@beardmann.dk>
> *Date: *Wednesday, June 2, 2021 at 08:46
> *To: *"toasters@teaparty.net" <toasters@teaparty.net>
> *Subject: *[EXTERNAL] NDMP Backup speeds
>
>
>
> Hi there
>
>
>
> We are doing NDMP backups from out FAS2750 with NL-SAS disks onto LTO8
> tape drives via NetBackup.
>
> For other reasons we are doing this via 10G Ethernet, and not directly
> connected tapedrives to the NetApp controllers.
>
> We are seeing backup speeds of about 40-50MB/sec. which is kinda slow, the
> backup is a Windows host that is doing nothing… no CPU or memory load as we
> backup.
>
> The LTO8 drives should be able to do much better (350MB/sec. before
> compression)…
>
> The aggregates are RAID-TEC with 17 drives in it. We are on ONTAP 9.7.
> We are not seeing a lot of load as we are running the backups…
>
> Could it be that the protocol itself it as cause here?
>
> In the “old days” we did NDMP backup directly from the controllers to the
> tape drives.
>
> We are now thinking of converting two 10G ports to fibrechannel ports… if
> possible…
>
> So I have two questions ????
>
>
>
> The FAS2750 is configured with RJ45 10G ports, but would it be possible to
> use FCoE and somehow connect this to a FC/Ethernet switch and be able to
> connect it to the tape drive which is 8G FC (LC Connectors) ?
>
> (I have not worked that much with FCoE… if the FAS2750 was configured with
> CNA-Ports, it would have been easy to convert that into FC)
>
>
>
> Has anyone tried to backup NL-SAS volumes similar to this over directly
> connected NDMP, and what kind of speeds are you seeing?
>
>
>
> Finally if anyone have any idea if the NetBackup setup can be tweaked
> somehow, that would also be useful… (we have tweaked it a bit)
>
>
>
> /Heino
> _______________________________________________
> Toasters mailing list
> Toasters@teaparty.net
> https://www.teaparty.net/mailman/listinfo/toasters