ADSM-L

Re: [ADSM-L] Is anyone getting > 100MB/s on a NetApp SM2Tape backup?

2012-04-27 14:19:41
Subject: Re: [ADSM-L] Is anyone getting > 100MB/s on a NetApp SM2Tape backup?
From: "Schaub, Steve" <steve_schaub AT BCBST DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Fri, 27 Apr 2012 14:17:32 -0400
Additional data

1.       The option we turned off that gave us our latest boost was "Options 
ndmpd.offset_map.enable off"

2.       I ran 4 concurrent SMTape backups to TSM yesterday evening. Total data 
throughput was 1,597,200 MB over 121 minutes, which works out to 220 MB/s or 
780 GB/hr. when I ran 4 concurrent backups (same volumes) with the above option 
on a few weeks ago, we got 83 MB/s or 292 Gb/hr.
So what we are seeing is some sort of per-thread NDMP limit.  Any ideas?
-steve

From: Schaub, Steve
Sent: Friday, April 27, 2012 9:50 AM
To: 'ADSM: Dist Stor Manager'
Subject: Is anyone getting > 100MB/s on a NetApp SM2Tape backup?

All,

We are implementing an IBM N-Series Filer cluster (rebranded NetApp) and 
working on getting the full-volume DR backups setup.  We are using 
SnapMirror2Tape with 3-way NDMP (don't want to dedicate FC tape drives from our 
Library Mgr environment, and we have 10gb Ethernet from the Filer to the TSM 
Server).   TSM Server 6.2.2 AIX on P-series.  Tape is IBM 3592, disk is IBM 
DS8k.

When we started testing, we were only getting about 60-80MB/s throughput.  
After turning off, we are now up to 90MB/s.  Filer cpu is < 40%.

We have run the following tests:
                NDMP Dump to null = 300-400MB/s (disk is not the bottleneck)
                NDMP Dump to the other Filer in the cluster = 104MB/s (takes 
TSM & tape out of the equation)
                NDMP Dump to another vol on the same Filer = 110MB/s (takes 
network out of the equation)

We're not getting much help from IBM Support, and at least one person has said 
that 100MB/s is about the best you can expect.

So I'm curious if anyone is getting more than that, and if not, what is holding 
it back?  I have at least one Windows 2008R2 SQL Server running under VMWare 
that has proven it can restore between 225-325MB/s on the same 10gb network, so 
I don't understand why we can't achieve better performance on a backup that is 
supposedly block level?

Thanks,

Steve Schaub
Systems Engineer II, Windows Backup/Recovery
BlueCross BlueShield of Tennessee

-----------------------------------------------------
Please see the following link for the BlueCross BlueShield of Tennessee E-mail 
disclaimer:  http://www.bcbst.com/email_disclaimer.shtm

<Prev in Thread] Current Thread [Next in Thread>