File transfers very slow
All,
I have a Ctera 800 box and a gigabit cisco sg 300. Cisco is set to auto, flow controll on, LAG on nics and switch ports, no jumbo frames, no port security. The Ctera box is set to auto and LAG. The data gets transfered to another server on the same switch and same server config except for LAG. The transfer rate never gets above 50 Mbps. Any suggestions.
Thanks
I have a Ctera 800 box and a gigabit cisco sg 300. Cisco is set to auto, flow controll on, LAG on nics and switch ports, no jumbo frames, no port security. The Ctera box is set to auto and LAG. The data gets transfered to another server on the same switch and same server config except for LAG. The transfer rate never gets above 50 Mbps. Any suggestions.
Thanks
Who is Participating?
First off I'm pretty sure that lag does nothing for a single file transfer but even then you should be able to hit 110-120MB/s on gigabit. You do mean Bytes and not bits right?
You will also be limite by the read speed o the source and the write speed of the destination. I'd star by eliminating read and write speed by transferring to/from a local drive (or a ram disk if you can)
If all that good, then look at the network.
You will also be limite by the read speed o the source and the write speed of the destination. I'd star by eliminating read and write speed by transferring to/from a local drive (or a ram disk if you can)
If all that good, then look at the network.
Aaron is correct about the LAG giving you no boost in speed. LAG's are great for failover and for increasing the total possible bandwidth, but that bandwidth boost only becomes apparent when multiple sessions are going across the LAG.
On a per-session basis, the algorithm used will choose which physical interface the data will traverse and that choice will never change throughout the session nor can it multiplex the session across multiple physical links. This means your maximum per-session speed can never exceed the speed of an individual link. Also, bear in mind that a 1G switch is 1Gigabit per second, not 1Gigabyte per second. Depending on which application you are using to display the speed of the transfer, it can make the transfer seem very fast... or unusually slow. A 1G switch only equates to 125Megabytes per second absolute max, and to bring that one step further the LAG will only allow a single session to hit 125MBps and this is without the overhead incurred with normal TCP/IP.
A LAG is still great in the fact that you can have multiple sessions going on that can add up to a total bandwidth of all available links, but individual sessions are limited.
The other thing that comes in to play with these situation is latency. To achieve the full gig throughput your latency would need to be .000524288s (.524ms) or less. Even on a direct connection this can be hard to achieve without the right hardware. When latency is getting this low, small latency differences cause big throughput differences. If you had .8ms latency you would have a max throughput of approximately 80MBps. 1.28ms latency would produce a max of 50MBps.
I realize the above is likely not to be your answer, but it's a good baseline to work off of.
On a per-session basis, the algorithm used will choose which physical interface the data will traverse and that choice will never change throughout the session nor can it multiplex the session across multiple physical links. This means your maximum per-session speed can never exceed the speed of an individual link. Also, bear in mind that a 1G switch is 1Gigabit per second, not 1Gigabyte per second. Depending on which application you are using to display the speed of the transfer, it can make the transfer seem very fast... or unusually slow. A 1G switch only equates to 125Megabytes per second absolute max, and to bring that one step further the LAG will only allow a single session to hit 125MBps and this is without the overhead incurred with normal TCP/IP.
A LAG is still great in the fact that you can have multiple sessions going on that can add up to a total bandwidth of all available links, but individual sessions are limited.
The other thing that comes in to play with these situation is latency. To achieve the full gig throughput your latency would need to be .000524288s (.524ms) or less. Even on a direct connection this can be hard to achieve without the right hardware. When latency is getting this low, small latency differences cause big throughput differences. If you had .8ms latency you would have a max throughput of approximately 80MBps. 1.28ms latency would produce a max of 50MBps.
I realize the above is likely not to be your answer, but it's a good baseline to work off of.
No I really mean bits. That is why I can never backup my server in a day. It would take weeks at this rate. We are looking into the database portion of the portal that the server is backing up to. Mind you they are both on the same LAN for now and not transfering over the WAN. Thanks all for the LAG knowledge. Do you think maybe the firmware on the switch? Thanks
Question has a verified solution.
Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.
Have a better answer? Share it in a comment.
All Courses
From novice to tech pro — start learning today.
-
Course of the Month7 days, 12 hours left to enrollMonitoring and Operating a Private Cloud with System Center 2012 R2 273 lessons
Remove the LAG and let the two machines communicate using a regular single connection. This removes/suggests the LAG as a culprit.
Connect the two machines directly together to remove the switch from the situation. This can remove/suggest the switch as a culprit.