Excellent article I have been struggling with the Multi-Pathing for a couple of weeks and this defiantly fix my slow 12Mbps transfer speeds I was getting. I am now getting maximum of 140MBps which is alot better but am I correcting in thinking I should be able to get more. If 1Gbps = 125MBps and I have two 1Gbps links two each controller with multi-pathing and round robin enabled I should be able to get 200-250MBps?

To get my transfer speeds I was using ATTO disk benchmark and Vsphere VM disk performance graphs

What sort of speeds should I settle for?

Regards
Dave Mac

Well 250MB/s is a theoretical limit that youy will never reach but yes if both NICs are pushed to a maximum you could get 200-210MB/s.

But there are several factors that can limit the throughput on your NICs. Size of the data being transmitted can make a huge difference in througput. 4K blocks of data causes more IOPS but lower MB/s, 32K blocks of data has fewer IOPS but more MB/s. Also, reads will usually push more data per second than writes. And of course sequential is faster than random (read or write) So with 32K blocks, 100% read, 0% random, with 2 NICs enabled you would probably get 170 - 190MB/s (roughly).

"Well 250MB/s is a theoretical limit that youy will never reach but yes if both NICs are pushed to a maximum you could get 200-210MB/s.

But there are several factors that can limit the throughput on your NICs. Size of the data being transmitted can make a huge difference in througput. 4K blocks of data causes more IOPS but lower MB/s, 32K blocks of data has fewer IOPS but more MB/s. Also, reads will usually push more data per second than writes. And of course sequential is faster than random (read or write) So with 32K blocks, 100% read, 0% random, with 2 NICs enabled you would probably get 170 - 190MB/s (roughly).
"

I agree with Aaron. You also have to take into account that your not lagging the ports. Your connecting through two seperate pipes, and then using a rule (RR) to say go down all paths and send data evenly. Other questions would include the use of jumbo frames and how the vSwitch was setup.

All and all if your getting 140 MB out the pipe, your doing really good.