• Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 3866
  • Last Modified:

LTO2 performance on Arcserve 11.5 and maximumSGList setting

Hi my situation:

Windows 2003 Standaard (SP1) - Arcserve 11.5  and LTO2 IBM Ultrium-TD2 (firmware 53y3) Powervault 136T and latest firmware....

what is the optimal Blocksize for the LTO2 drives.  i think 256KB.

now i found this article on the microsoft site :  http://support.microsoft.com/kb/907418/EN-US

it is about performance limitation after installing SP1 on windows 2003 server. it is about  the tape.sys.

my question:  does this fix have any effect on arcserve, because all devices are disabled in system manager even the media changer. and the fix is for Tape.sys  (i think meant for NTBACKUP)

how can i tune my server for high(read better) backup performance.  my average backup speednow is 500-1100Mb/Min, and when i calculate, native for LTO2 (IBM) = 35MB/sec = 2100MB/min  

I had DLT 8000  with 6MB/Sec native, my backup speed on exactly the same server was 350 MB/Sec.

what can i do about is..

can i do some registry change like i did for DLT for this LTO2 issue.

what are your experiences......

(sorry for my bad english..)

Greetz charly  
0
CharlyD
Asked:
CharlyD
  • 7
  • 5
2 Solutions
 
CharlyDAuthor Commented:
this is what i found at the HP site...

The block size of your tape drives is larger than 64 kb. This is the default upper limit for Windows.

You can tweak it however in the registry of Windows to use a larger block size:

"With the new equipment with better performance, used in a SAN environment, one can configure the tape device with higher block size. In order to improve the backup performance, for LTO or STK 98840 drives, it's common to have block size equal to 256k.

However, on windows media agent hosts, there might be a limitation due to the HBA card. Some of them are not able to process more than 64k. The MaximumSGList parameter should be modified in the registry entry. This parameter is located in: HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\<your_card>\Parameters\Device\
MaximumSGList = (DesiredBlockSize/4) + 3


Example with a Compaq HBA card.

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Servic
es\CPQKGPSA\Parameters\Device\MaximumSGList= 67"


my card is a 39160  ....

answers needed..
0
 
dovidmichelCommented:
As far as I know 64k is the limit and it is set by the controller as per the limitation in Windows, so no help there.

Update ARCserve
http://supportconnectw.ca.com/public/storage/downloads/nt/115/basbr115-solpatch.asp
QO77091 ZIP 492 MB 03/22/06
QO78909 ZIP 13.1 MB 05/10/06 NT -DEVICE SUPPORT UPDATE 5

(IBM) = 35MB/sec = 2100MB/min   - thats nice but more like fanticy land and not real life. From what I've seen your throughput range is in the norm, and that spread is probably due to difference in data (lots small files = slow, large files = fast).

Start looking at data source and path. If this is all local it makes it a lot easier, if there are systems being backed up over the network also it becomes more complicated.

Basically the LTO2 & 3 drives reach harddrive speed and so there is a good chance the bottleneck is not at the tape drive. Use Windows Performance Monitor or some other tools to measure the raw throughput.

As a test stop the ARCserve Tape Engine, open a command box and go to the ARCserve directory, run "simulate -ca". It will let you simulate a tape drive and do a backup. Now the data just gets dumped into a virtual trash can and you can see if the throughput is any better than when the real tape drive is used. Be sure to run simulate again to remove it when through.
0
 
CharlyDAuthor Commented:
but i did cstop

then i ran  "simulate -ca"

i configured it and then.. cstart and go.. or??
0
Concerto Cloud for Software Providers & ISVs

Can Concerto Cloud Services help you focus on evolving your application offerings, while delivering the best cloud experience to your customers? From DevOps to revenue models and customer support, the answer is yes!

Learn how Concerto can help you.

 
dovidmichelCommented:
well actually just the Tape Engine needs to be stopped.

yes, start it up and run a backup of the same data.
0
 
CharlyDAuthor Commented:
hi what is the exact procedure can t get it to work...
0
 
IPKON_NetworksCommented:
This is such a difficult, and routinely asked, question. The main issue relates to so many factors that I shall try and list them out.

1. Source data
A large part of transfer rates relates to the source data. Large block files will transfer quicker than lots of small files. Setting the block size to 64Kb (for example) will be great for files that are multiples of 64Kb. However, if you have 1000's of 3Kb files then this will not necessarily help you.

2. Disk location
Is the data located on local DAS disks or on a SAN or across the network. If the I/O is having to be processed by the server prior to being sent to the backup server, just to be processed again, then this slows down the transfer rate. What you see is the end result (ie, rate to tape). What you NEED to know is the various rates between each stage of the process.

3. Bus Speeds
How many controllers are connected to the tapes? Are you feeding the tapes fully to utilise their bandwidth? A common error is that people assume that LTO can perform at 35MB/s therefore why am I only getting 10MB/s. Well, LTO drives are 'clever' in that they can perform in steps depending on the input speeds. (ie how much data you can feed them /sec) However, there is a lower limit. If it is not getting enough data in it will step down to a default speed. On LTO2 this is 8MB/s I believe. It might even be lower.

4. Network etc...
How is the data getting to your backup server and the tape drive? Is it coming over a 100MB network from another server? Is this congested already? Do you have access to a dedicated Gigabit network for the backup? If so, are the NIC's on source and backup server on seperate PCI bus from the disk/tape controllers? What speed is the PCI bus?

All these things factor into the way the data gets onto the tape. If you think shortest path all the way (including internal server bus etc) then you can't go far wrong. Ultimately there is always a bottleneck and our challenge is to make that bottleneck the tape drive itself and nothing else.

Hope this helps
Barny
0
 
dovidmichelCommented:
after created an imaginary tape drive with simulate just start the tape engine and setup a job. You might have to go to the arcserve device manage window and format the fake tape in the fake drive before it will backup to it.
0
 
CharlyDAuthor Commented:
but i still don t know which steps to do after each other, i have the most simple test config,  server local disk and 39160 with 3 LTO2 drive and the lib. on a 2940UW.

ok 1. stop tape engine.
    2. simulate -ca  (in arcserve dir )
    3.   ...
    4.   ...
    5.   .
0
 
dovidmichelCommented:
stop tape engine.
simulate -ca
pick item 1. Add Simulated Devices
Number of Boards to simulate(Default to 1)?
enter for default
Tape size in data blocks( 16KB/block )? Default to infinite :
enter for default
Early warning position in blocks( 16KB/block )? Default to 0 :
enter for default
Write tape data to file? (y/n) Default to (NO)
enter for default
 >>>Begin to setting up board 0
Number of Stand Alone SCSI Tapes on Board:0 (Default to 0)?
enter 1
Add Changer to board? (y/n) Default to (Yes)
enter n

then it will ask for confirmation and you are ready to test it. Device config will now show only the one simulated tape drive. format a tape in it and then run a backup to it. Backup should be of a good size say around 50gb at least, and remember small files = slow, large files = fast.

After the test backup, go back to simulate and select to remove the simulated drive. Now run Device Configuration to setup the library again.

Still like I said before I doubt the bottleneck is at the tape drive. That is unless you are doing three backups at the same time then the SCSI bus is probably getting saturated and it would be worth thinking about adding a second controller. General rule for high end tape drives is no more that two drives per bus.
0
 
CharlyDAuthor Commented:
total directorys             398
total files                     6.292
total skips                    0
total size                      2.246,08 MB
total media                   2.267,37 MB
Elapsed time                 5m 8s
Average Througput        441,69 MB/min


second test was much better

total directorys                     1
total files                             15
total size disk                       14.919,87 MB
total size media                    23.544,00 MB
Elapsed Time                       14m 19s
Average troughput                1.644,51 MB/Min

this is the throughput i want but the top one is bad in my vision  What can this be ??

they were both made on the local server the second ons is from a software (IDE) raid 5 the first one from a normal 4gb disk scsi

could it be the disk performance?? arcserve is also installed on the 4GB Scsi disk. ( Compaq ST34572W )

the raid 5 is build from 4x WD80 0JB-00CRA1  for the second (fast ) backup job

could is be the disk performance??

or were do i have to look.





 
0
 
dovidmichelCommented:
top one is small files and so slower, next one show larger files and so higher throughput.

So instead of 1,100mb/m it got 1,600mb/m which is better.

so as mentioned three drives on the one controller can be too much, two drivers per will help.

Beyond that the bottleneck can be just about anyplace from the hard drives to the tape drives, could even be the PCI bus that is maxing out. That can be checked with the manufacturer of the system.

but this shows the greatest difference is caused by the data itself and there is no way around that, lots of small files is a lot of overhead which is going to slow things down.

Performance Monitor is a good tool to use and there are others. If you ck into the configuration specs for these great through put claims you find out the equipment used is almost outlandish it is such high end stuff. Now talking real world your thoughput is in the normal range.

But like I said one place to start is adding another controller, have one going to two drive and the other to one drive and the library. Next place to start is the other end. Check the firmware on all the hard drives for the RAID they should be on the same latest and greatest revision.

A second reason (beyond the small files) for the backup of the SCSI drive being slower is that it is also running ARCserve so it has to do both read the data and handle to overhead from ARCserve.

Now get back to simulate if you have not done it already and remove the simulated drive and then run Device config to setup the library again and get you normal backups done.
0
 
dovidmichelCommented:
another thing came to mind.

The job log is handy for this type of thing because it gives clear relivant data, volume, # dir, # files, time, throughput.

I have found that taking the time to make a chart of this data realy helps to get a clear idea of just what is going on.  
0
 
dovidmichelCommented:
tough to say, the suggestions I gave helped others with throughput problems but there are so many things that can effect it unless CharlyD comes updates I don't know if I helped her or not.
0

Featured Post

Get your Conversational Ransomware Defense e‑book

This e-book gives you an insight into the ransomware threat and reviews the fundamentals of top-notch ransomware preparedness and recovery. To help you protect yourself and your organization. The initial infection may be inevitable, so the best protection is to be fully prepared.

  • 7
  • 5
Tackle projects and never again get stuck behind a technical roadblock.
Join Now