cancel
Showing results for 
Search instead for 
Did you mean: 

Slow Throughput LTO-Ultrium 2

AndyTh
Level 3
 Hello,

I am only getting 12gb/hr throughput on my new Quantum LTO-Ultrium 2 drive. This drive should theoretically get 96GB/hr, so it is way under. I am expecting somewhere around 80GB/hr.

Here is my setup:
Quantum LTO-Ultrium 2 Half-Height drive.
Adaptec 29160 with the latest driver and Firmware.
Veritas 9.1 4691. I have attempted to apply SP4 and SP4a for Veritas, but I am not sure that it was successfully installed. It doesn't show up when you got to About Backup Exec...
OS is Windows XP SP2

I haven't changed anything for the drive properties in terms of Block Size or cache. The block size is currently 64k.

The throughput is roughly the same for local files as it is when backing files across the network.

The drive is terminated.

Any ideas?

Andy
10 REPLIES 10

lyon
Level 5
Dumb one but I've seen it before - remove the terminator.
Also check the SCSI card setup.
What hard drive are you using? If you are on XP, is it a standard IDE drive?

Lyon

AndyTh
Level 3
 The backup is running right now, but I will check the SCSI card setup. What, in particular, should I look for?

Andy

lyon
Level 5
 Is it set for 40mb/sec or something silly. You've got a 29160 on a standard PCI bus or PCI-X? What backplane? This could be a disk or backplane bottleneck

Lyon

AndyTh
Level 3
 It is a standard PCI Bus. This is running on a Generic "MWAVE" Workstation with an ASUS P4P800-VM motherboard

Andy

lyon
Level 5
 Standard PCI architecture and an IDE-100 drive will give you up to 45GB/hr maximum. With Windows running and the usual background services, you can bring that down a lot.

To increase the throughput you need to move the data to its own disk with a faster IO and a better machine.

Lyon

AndyTh
Level 3
 Thanks for your replies, Lyon. I checked the SCSI BIOS settings, and everything is set for 160MB/Sec. The Hard Drive is IDE-100. Do you have any other suggestions besides upgrading to a better machine? Resources are a bit limited. I may be able to move the backups to a server machine with SCSI drives, but I was hoping to keep the backups running on a separate machine. Thanks again

Andy

teiva-boy
Level 6
 PCI is insufficient to reach maximum bandwidth for LTO2 or 3 and most def 4 and 5.

You need to be running PCI-X or better, preferably PCI-E so that it has it's own PCI bus that is not shared with other devices on the motherboard.

PCI, IDE, USB1.1...  Might as well be an ISA SCSI card connected through a 50pin centronics connector..

 Time for an upgraded PC and SCSI card, period.  



THarry
Level 5
Just to continue what Lyon said:

One of my media servers is a dell gx270 plain-jane pc,IDE 100, 29160 card on Standard PCI, but I'm running Netware 6.5 on it.

My jobs are averaging 716Mb/min or almost 43Gb/hr. However, I have one volume of 14Gb that runs at 1280Mb/Min (76.8Gb/hr) and another volume of 33Gb that runs at 1029Mb/min (61.7Gb/hr). All data is remote.

No real point here, just some real-life data

Harry

teiva-boy
Level 6

Both the GX270 and the Asus motherboard mentioned thus far, share the same Intel chipset, but not the same OS.  Frankly, Netware is much more efficient than Windows XP IMO.  Interrupt handling, file server performance, networking performance, all better on Netware (but this isn't a debate on why Netware was so great ;) )


But there are some subtle hardware differences too (taken from respective manufacturer’s links)
The Asus MB only has a 10/100 NIC, Dell a 10/100/1000Mb NIC (weird though as the Intel reference board is 1Gb?)
Asus has 3 PCI slots and one AGP, Dell varies by form factor; one, up to three on the full desktop model


Now, that said, looking at the Intel 865G chipset specs, the biggest achilles heel that I can see is the southbridge, the ICH5.  Man that thing is taxed!  The southbridge must handle all of the following devices, process them and allocate enough bandwidth to them.

  • USB ports
  • PCI slots
  • IDE
  • Sound
  • LAN


Bandwidth from southbridge to northbridge is a paltry 266MB/sec.  Running a Gb nic at 80% capacity, two IDE hard drives (OS and data drive), and an UW SCSI card WILL saturate the south to northbridge link.  

Now back on topic...  96GB/hr is 1.6GB/min is 27MB/sec (rounded off)  All assuming MAX theoretical speed.  

No IDE hard drive can do 27MB/s of sustained throughput throughout a backup job.  RAID a few drives together, and it’s possible.  A 100Mb NIC will only get some 12MB/s MAX in a fairy world, realistically 8MB/s (same goes for Gb links, 80MB/s is about the real world max). 

BTW, did you know that LTO-2 needs a MINIMUM of 30MB/s to maintain sustained writing to tape?  It can write faster with compression and files that are compress’able, up to 4x faster.  But I normally only ever see some 1.0-1.2 ratios in my BE environments.  Thus an average of 36MB/sec for me…  Cant spool 36MB/s to the drive, fine, the tape will rewind, buffers will fill, the tape will fast forward, and continue writing where it left off.  It’s called tape shoe-shining.  That’s why so many people that go to LTO-3 and  LTO-4 find their backup times actually increased, or don’t change at all, and they get mad and blame the software, because it cant be their new 10k tape drive…

 

Ultimately what I’m saying, hardware wise and OS wise, it a large leap to assume you’ll get the throughput needed out of your computer to meet even the minimum throughput levels for LTO-2.  You have too many variables already against you.  The chipset, apparently the NIC, and the OS.

 

Backup systems are a great way to find bottlenecks in your servers, your network, and your disk subsystems.  It’s easy to buy the latest and greatest servers, 10K SAS drives, only then to cripple it with incorrect RAID array’s, and a cheap Gb switch with shared buffers for all 24 ports… 

tftact
Not applicable

Hi Andy,

I have exactly the same throughput on a similar system, but with another software.  By chance have you solved your problem ?
I have an HP LTO-2 drive and I have used the HP performance programs, which showed exactly the same performance (about 3-4 MB/s.

Michel.