We are using Legato Networker 5.5.1 build 115 and are also experinceing poor performance we have an atm connection to a local backup and the speed is only reaching 4 mb/s the norm for all of the drives in the silo is approx 7-9 mb/s and we are using DLT7000 drives any help would be greatly appreciated.
Tek-Tips is the largest IT community on the Internet today!
Members share and learn making Tek-Tips Forums the best source of peer-reviewed technical information on the Internet!
-
Congratulations strongm on being selected by the Tek-Tips community for having the most helpful posts in the forums last week. Way to Go!
POOR PERFORMANCE
- Thread starter brsluss
- Start date
brsluss,
Have you attempted to set your network adapters to 100MB/s Full Duplex and you should force the appropriate ports on your switch to reply the same.
NIC on ServerA = 100MB/s Full Duplex
Port 22 (ServerA) on Switch = 100MB/s Full Duplex
If this is a server with 1000's of small files, for example a Session Wall, you may desire to set the NIC and Port to 100MB/s Half Duplex, due to the amount of overhead for small files.
I can't seem to get our session wall into the server at speeds of greater than 6MB/s.
Hope this helps,
CF
Have you attempted to set your network adapters to 100MB/s Full Duplex and you should force the appropriate ports on your switch to reply the same.
NIC on ServerA = 100MB/s Full Duplex
Port 22 (ServerA) on Switch = 100MB/s Full Duplex
If this is a server with 1000's of small files, for example a Session Wall, you may desire to set the NIC and Port to 100MB/s Half Duplex, due to the amount of overhead for small files.
I can't seem to get our session wall into the server at speeds of greater than 6MB/s.
Hope this helps,
CF
What I would do is:
1) Check to see what type of ATM connection you have: OC3...
2) Check Memory/CPU usage & network bandwidth.
3) Backup a local filesystem & note the speed.
4) If all others are okay:
a) alot of available bandwidth.
b) low memory/cpu usage.
c) Large ATM pipe.
d) Local filesystem is backing up at alot higher speed.
I would increase paralellism & break-up large filesystems into directories. I got twice the performance by doing this. Networker uses one stream per filesystem.
1) Check to see what type of ATM connection you have: OC3...
2) Check Memory/CPU usage & network bandwidth.
3) Backup a local filesystem & note the speed.
4) If all others are okay:
a) alot of available bandwidth.
b) low memory/cpu usage.
c) Large ATM pipe.
d) Local filesystem is backing up at alot higher speed.
I would increase paralellism & break-up large filesystems into directories. I got twice the performance by doing this. Networker uses one stream per filesystem.
Guest_imported
New member
- Jan 1, 1970
- 0
Hi,
Your problem has nothing to do with network hardware performance. What you need to do is bump your parallelism up to 35 in your server setup (512 max in cluster environment with multiple storage nodes).
Also adjust your Tape setting so that it's max sessions is 35. You will then be multiplexing your data in up to 35 streams at once onto a single tape.
Remember, each client entry on the server has a setting for parallelism too. (Default is 4 so the REPAIR DISK, SYSTEM STATE, Drive C and Drive D will all backup simultaneously.
C A U T I O N: If you have multiple partitions on 1 drive, you will get thrashing on that drive if they are all backing up at once. If you can live with that, leave the client's Parallelism at 4 or more, otherwise bump it down to 1 or 2.)
I can backup 26 GB from 9 NT servers and 1 NT workstation in 3 hours on a 100 Mb/s ethernet network.
Now if you are using multiple drives, you'll get even better performance than I do but restores will take a lot more time as each tape drive has to search for a client's thread to read.
Good Luck !!!!
(John.Trembly@dpcdsb.org)
P.S. I'm just setting up our backup server and now I'm testing total NT system recovery and getting an error that says I can't overwrite secured files. Legato is definately not for the faint hearted !!!
Your problem has nothing to do with network hardware performance. What you need to do is bump your parallelism up to 35 in your server setup (512 max in cluster environment with multiple storage nodes).
Also adjust your Tape setting so that it's max sessions is 35. You will then be multiplexing your data in up to 35 streams at once onto a single tape.
Remember, each client entry on the server has a setting for parallelism too. (Default is 4 so the REPAIR DISK, SYSTEM STATE, Drive C and Drive D will all backup simultaneously.
C A U T I O N: If you have multiple partitions on 1 drive, you will get thrashing on that drive if they are all backing up at once. If you can live with that, leave the client's Parallelism at 4 or more, otherwise bump it down to 1 or 2.)
I can backup 26 GB from 9 NT servers and 1 NT workstation in 3 hours on a 100 Mb/s ethernet network.
Now if you are using multiple drives, you'll get even better performance than I do but restores will take a lot more time as each tape drive has to search for a client's thread to read.
Good Luck !!!!
(John.Trembly@dpcdsb.org)
P.S. I'm just setting up our backup server and now I'm testing total NT system recovery and getting an error that says I can't overwrite secured files. Legato is definately not for the faint hearted !!!
Just a note: Every action has a consequence. If you increase your target session on a single tape to 35, and you are in an evironment that has multiple drives, then you are telling the server to send 35 streams of data to drive no. 1 before sending anything to drive no. 2. The idea is get all your drive spinning as quickly as possible.
Also, The more streams stripped across a single tape/drive, the faster the backup. However, the more savestreams stripped across the tape/singel drive translates to slower recoveries.
When tuning the settings of your server, keep these things in mind. Target sessions on a drive is not an absolute configuration parameter. Sever & Client Parallelism are absolute and limiting. This means that having a low target session will not restrict the amount of savestreams that can be written to tape, it merely is a prefernce which helpes balance the amount of data stripped to each drive.
Hope this helps.
mkeane
Also, The more streams stripped across a single tape/drive, the faster the backup. However, the more savestreams stripped across the tape/singel drive translates to slower recoveries.
When tuning the settings of your server, keep these things in mind. Target sessions on a drive is not an absolute configuration parameter. Sever & Client Parallelism are absolute and limiting. This means that having a low target session will not restrict the amount of savestreams that can be written to tape, it merely is a prefernce which helpes balance the amount of data stripped to each drive.
Hope this helps.
mkeane
Hi,
first I configured networker for a good backup performance. Set the parallelism to a high value and the target session too. Backing up our clients works very well. But the result was, that the recovery takes a lot of time (the tape reads the data about 5-7 Mb/s -> writing to disk under 1 Mb/s).
Because I think recovery-time is more important than the time a backup takes, I want to re-configure networker so it doesn't use multiplexing. Setting only the target sessions to 1 won't work, because the server-parallelism is decisive!
What can I do to completly prevent multiplexing???
first I configured networker for a good backup performance. Set the parallelism to a high value and the target session too. Backing up our clients works very well. But the result was, that the recovery takes a lot of time (the tape reads the data about 5-7 Mb/s -> writing to disk under 1 Mb/s).
Because I think recovery-time is more important than the time a backup takes, I want to re-configure networker so it doesn't use multiplexing. Setting only the target sessions to 1 won't work, because the server-parallelism is decisive!
What can I do to completly prevent multiplexing???
You may not want to completely disable multiplexing. You simply need to find that balance that works for you. A target session of 4 for each drive is the suggested amount of data that can keep the drive spinning at an optimal speed. I would suggest setting the target session to 4 and then setting the server parallelism to # drives x traget sessions.
Server Parallelism= # drives x Targe Sessions
Hope this helps.
mkeane
Server Parallelism= # drives x Targe Sessions
Hope this helps.
mkeane
After optimizing your Network, i.e 100 MB full, GB NIC etc, look at your paralism setings for Networker server, having paralism settings too high could cause more problem then you can think.
I backup 86 servers with Proliant 3000 Dual GB card, Most heavy servers on GB NIC on Backup Network 600 GB incermental and 1.5 TB full with 3 DLT7000 drives.
Paralism at 24, Drive sessions to 6.
DLT drives funtion beter if you can supply them constant Data from Multiple server at the same time.
Try using Two client for your biggest server with Breaking up the savesets half and half.
I use the above to backup Partitioned Notes server over 250 GB total data.
Two cleint for the same sever will allow you schuduling\backing up flexibility.
I backup 86 servers with Proliant 3000 Dual GB card, Most heavy servers on GB NIC on Backup Network 600 GB incermental and 1.5 TB full with 3 DLT7000 drives.
Paralism at 24, Drive sessions to 6.
DLT drives funtion beter if you can supply them constant Data from Multiple server at the same time.
Try using Two client for your biggest server with Breaking up the savesets half and half.
I use the above to backup Partitioned Notes server over 250 GB total data.
Two cleint for the same sever will allow you schuduling\backing up flexibility.
The thing to remember about streaming tape (DLT) drives is that they work by moving the tape rapidly over the heads. If you do not present enough information to the tape drive, then it must stop, rewind, and accelerate back to speed - this really slows down your backup! For our site, we find a minimum of 8 parallel sessions are needed to keep our tapes busy - mainly because we have large numbers of small files.
What type of ATM card are you using? What is your o/s platform?
Check if the ATM NIC uses multiple NDIS receive buffers. Currently Microsoft's TCPIP stack does not support the use of these chained MDLs on receive.
Does the problem happen if you are using Ethernet instead of ATM? This will narrow down the problem to whether it is a network issue or a NetWorker issue.
Check if the ATM NIC uses multiple NDIS receive buffers. Currently Microsoft's TCPIP stack does not support the use of these chained MDLs on receive.
Does the problem happen if you are using Ethernet instead of ATM? This will narrow down the problem to whether it is a network issue or a NetWorker issue.
Hi,
I have found that a target session of 3 is the ultimate setting. This must be in relation to the server parallelism
i.e targetsessions * number of tape drives = server parallelism. With this setting you will get almost the same backup and recovery times, using DLT700 or AIT-2.
If you completey want's to avoid multiplexing I would suggest that you use a diskbackup solution and then stage the savesets to tape.
We use this technic on two of our backup servers, backing up over 1 TB a day. This way we always stream the drives and clients won't have to wait for tapes. Of course the server is running almost 24 hours, but it's not realy a problem.
If I where you, I would also consider upgrading Legato, going from 5.5.3 to 6.1.2 gave us a real boost.
Best Regards
Staffan
I have found that a target session of 3 is the ultimate setting. This must be in relation to the server parallelism
i.e targetsessions * number of tape drives = server parallelism. With this setting you will get almost the same backup and recovery times, using DLT700 or AIT-2.
If you completey want's to avoid multiplexing I would suggest that you use a diskbackup solution and then stage the savesets to tape.
We use this technic on two of our backup servers, backing up over 1 TB a day. This way we always stream the drives and clients won't have to wait for tapes. Of course the server is running almost 24 hours, but it's not realy a problem.
If I where you, I would also consider upgrading Legato, going from 5.5.3 to 6.1.2 gave us a real boost.
Best Regards
Staffan
- Status
Similar threads
- Locked
- Question