Try Secure Copy (the demo is a useless). Draconian licensing, but you can add and remove servers from the license. We are migrating 20TB of small tif, txt, etc and this is the best tool we have found. We put in on an old backup server with a gig link and copy from our Windows Cluster to Celerra. You tell it how many threads to run and it takes care of setting up that many copy sessions and keeps track of them. Can fill the pipe with 8 threads and >100MB files. Ran 18 threads with <4K files and got about 50% NIC utilization. No time or patience to try to do this with Robocopy. Probably get higher perf if you run it directly on the source server, but we haven't tried that on the user file clusters out of fear...
From the help file: By default, the maximum number of files in a batch is 25 and the maximum size of the batch is 1MB. The minimum number of threads is 2 by default. By increasing the thread count, you are adding more threads that copy batches of files. Each of the file threads deals with one batch of files at a time.
In normal operation, these default settings should be adequate. However, if you have a large number of small files and many batches, you might start by creating at least 10 threads, and then monitoring network card utilization.
The goal of adjusting these three parameters is to maximize your network bandwidth. Other factors that prevent Secure Copy from using all the network bandwidth include slow disk speeds, disk fragmentation, and other activity on the network and file server.
Check exactly what you are migrating. Small files take longer than large ones; by small I mean less than 8K. Anything less than 8K doesn't fill network IP packets as efficiently as larger ones, so it takes longer.
It's like having to moving a pile of rocks one at a time by size. The more small rocks you have, the more times you go back and forth with your wheelbarrow. The more big rocks, the fewer trips.
From the CLI do "server_netstat server_2 -i | grep cge" several times to see what your traffic really looks like. Is it balanced? Does one interface get more traffic than another? A properly balanced set of nics (ex 4 trunked with etherchannel) will have two handling outbound packets and two handling inbound. There will be a small amount of acks returning out each channel, but it will be obvious if there is something out of whack. Check the server_netstat against what your router says.
Network imbalance is possible, but I'd put my money on a lot of small files.
dynamox
9 Legend
•
20.4K Posts
0
September 15th, 2009 12:00
robocopy Y:\ Z:\ /MIR /NFL /FFT /R:1 /W:1 /NP /COPYALL /B /LOG+:C:\robolog.txt
Disk Jockey
1 Rookie
•
92 Posts
0
September 18th, 2009 08:00
From the help file:
By default, the maximum number of files in a batch is 25 and the maximum size of the batch is 1MB. The minimum number of threads is 2 by default. By increasing the thread count, you are adding more threads that copy batches of files. Each of the file threads deals with one batch of files at a time.
In normal operation, these default settings should be adequate. However, if you have a large number of small files and many batches, you might start by creating at least 10 threads, and then monitoring network card utilization.
The goal of adjusting these three parameters is to maximize your network bandwidth. Other factors that prevent Secure Copy from using all the network bandwidth include slow disk speeds, disk fragmentation, and other activity on the network and file server.
Tripper3
1 Message
0
September 18th, 2009 08:00
It's like having to moving a pile of rocks one at a time by size. The more small rocks you have, the more times you go back and forth with your wheelbarrow. The more big rocks, the fewer trips.
From the CLI do "server_netstat server_2 -i | grep cge" several times to see what your traffic really looks like. Is it balanced? Does one interface get more traffic than another? A properly balanced set of nics (ex 4 trunked with etherchannel) will have two handling outbound packets and two handling inbound. There will be a small amount of acks returning out each channel, but it will be obvious if there is something out of whack. Check the server_netstat against what your router says.
Network imbalance is possible, but I'd put my money on a lot of small files.