Hey guys, I run my main computers off my gigabit switch. This includes my primary computer, file server, secondary, tertiary, and my PBX server. Currently, only the primary, secondary, and file server have gigabit ethernet ports, so for the rest of this thread, assume I'm only speaking of transfers between those three. My router, which is my current DHCP and DNS provider, is 10/100 mbps. If I try to do any file transfers between the three gigabit connected computers, the transfers are capped at 30MBps. Does this suggest the router is causing issues? If I were to set up a DHCP server with gigabit ethernet, would this eliminate that problem?
Thanks
-Neil
Last edited by ntomsheck : 07-26-2010 at 05:51 PM.
Do you mean 30 megabits per second ( Just under 4 megabytes a second ) or 30 megabytes a second ( 240 megabits/second ) ?
On my mostly gigabit network the 'fast' transfers cap out at 30 megabytes a second. When I benchmark from ram drives I can hit 60 megabytes/second ( 480 megabits/second ). The theoretical maximum speed of a gigabit network is 125 megabytes/second, getting a quarter of that is normal unless you're transferring from solid state disk RAID arrays through a very high end system to a system of the same type.
I do have a gateway with a gigabit card in it facing my network. The only non-gigabit items on my network are a WAP and a printer. I can't remember what it's called, but there is cache on your switch to help deal with fluctuating speeds. Jumbo frames - which would help you get closer to the theoretical maximum speed - will not work well/at all when there are non-gigabit items on your network.
Hey guys, I run my main computers off my gigabit switch. This includes my primary computer, file server, secondary, tertiary, and my PBX server. Currently, only the primary, secondary, and file server have gigabit ethernet ports, so for the rest of this thread, assume I'm only speaking of transfers between those three. My router, which is my current DHCP and DNS provider, is 10/100 mbps. If I try to do any file transfers between the three gigabit connected computers, the transfers are capped at 30MBps. Does this suggest the router is causing issues? If I were to set up a DHCP server with gigabit ethernet, would this eliminate that problem?
Thanks
-Neil
Sounds like something doesn't support Jumbo Frames or that Jumbo Frames hasn't been configured/enabled on the relevant hardware. If your Gigabit switch and the servers all support Jumbo Frames (preferably up to 9KB), you should be able get significantly higher transfer rates. I transfer data on my network between 60 and 80MBps, depending on the type of data.
Can you let us know the make, model and revision of the switch you're using? Also, in the properties of the network adaptors (via Device Manager), you may need to enable Jumbo Frames and configure the frame size. If there's no Jumbo Frames option, you should see if there are later drivers for those network adaptors. Often later drivers add the ability to manipulate Jumbo Frames settings.
Never had speed issues mixing Gigabit hardware with Fast Ethernet stuff on the same network, but from what I've read, this can happen when flow control is in use.
30MBps as in megabytes per second. I use the standard B byte, b=bit. Hopefully that clears any confusion.
The gigabit switch is a DGS-1008D by d-link, hardware revision C1
With jumbo frames set to 9kb, my transfers cap at roughly 2.5MBps. The file server can support up to 16kb, but my primary only supports up to 9. I would assume this means the switch does not support jumbo frames?
If I were to invest in a switch that supports super jumbo frames, would it cause network errors if the server was set to 16kb, but everything else to 9? (that supports it).
Last edited by ntomsheck : 07-28-2010 at 08:09 PM.
30MBps as in megabytes per second. I use the standard B byte, b=bit. Hopefully that clears any confusion.
The gigabit switch is a DGS-1008D by d-link, hardware revision C1
With jumbo frames set to 9kb, my transfers cap at roughly 2.5MBps. The file server can support up to 16kb, but my primary only supports up to 9. I would assume this means the switch does not support jumbo frames?
If I were to invest in a switch that supports super jumbo frames, would it cause network errors if the server was set to 16kb, but everything else to 9? (that supports it).
Looks from here that it does support 9KB jumbo frames. There are a coupla guys in here who know more about this stuff than I do. Will PM them and see if they can shed any more light.
Also - how far apart is each device, what sort of cabling are you using (CAT5e/6/6a) and have you tried swapping the cables?
Some new information:
With both machines set to 9kb frame size (now that I know the switch supports jumbo frames). I managed to get only 17-20MBps transfers (very inconsistent, I know). That's with QoS disabled. I noticed flow control was turned on, on my server, so I disabled it. The other machine used for the test already has it disabled
-With flow control disabled on the server, no transfers happen.
-With flow control disabled, and with QoS enabled, no transfers happen.
-With flow control enabled on both machines, neither computer can communicate with each other.
-With flow control re-enabled on the server only, and QoS enabled, transfers are down to 2-3MBps
-With flow control re-enabled on the server only, and QoS disabled, transfers are back up to 17-20MBps (back where I started)
By the way, the disk transfer rates are as follows:
Primary computer: avg 60MB/s
Server OS harddrive: avg 45MB/s
Server storage drive: avg 70MB/s
The files are being transfered to and from the storage drive. Even if it invokes the operating system harddrive, it should at least cap the transfers around 40 or 45MB/s. I assure you, that is the slowest component of the whole test circuit.
Last edited by ntomsheck : 07-30-2010 at 12:52 AM.
1.) Are the server's nics set to 9K jumbo frames?
2.) What are you using to measure transfer rates?
3.) With QoS, have you check that all devices are using the same QoS settings?
4.) Are there multiple active connections to the servers when the slow downs are occuring?
Well, when I don't specify a certain machine, I'm stating that it is applied globally. ex:
QoS is disabled - This means it is disabled on both machines used in the test circuit
QoS is disabled on the server - This means it is only disabled on the server.
My appologies for the confusion. For measuring network transfer rates, I'm taking a large (1.5GB) file and transfering it to the server. My primary computer has windows 7, which states the current transfer rate. I have already checked to see whether that method is at least in the ball park by timing the transfer of a known file size, and comparing it to the average transfer rate stated. It's accurate in my tests.
Yes, when I state I set the frame size to 9k, I'm speaking for both machines.
There are no other active connections while I test this. I disabled the FTP console and apache for these tests.
Last edited by ntomsheck : 07-30-2010 at 10:31 AM.
Ping with anything less than 9000 returns a "request timed out". At 9000, it says the packet must be fragmented. Now we're getting somewhere, but where? Both end points are configured for jumbo frames and according to D-link, the switch supports jumbo frames. I'm going to unplug everything but the two end points and set them to a static IP, I'll post back with results.
As for jperf, my only linux box has a 10/100 The gigabit equipped machines are all running windows. I might consider locating one of my live CDs if the the test returns nothing. Will jperf run from a flash drive in conjunction with a live CD?
Okay so with 9k frames set globally, only two things connected to my switch, and the IP address set manually, I'm able to ping up to 1490 bytes. Not one more. Anything above 1490 says request timed out. If I ping 9000 bytes however, it returns with 'file must be fragmented'. If I disable and/or set packet size to 1500 globally, I can ping any value under the sun. I also tried this on 4500 bytes and could not find a successful value above 1490 bytes. I did notice an anomoly though - when selecting packet choices, machine A has the choices of 1514, 2514, 4514, and 9014 bytes. The server has the choices of disabled (I assume 1500 standard), 4088, 9014, and 16128 bytes. I understand some manufacturers dont standardize the 4 byte header, but those values don't seem to agree with eachother, sans 9014.
Windrunner, to answer your question - the devices which do not support as high as 9kb would send a request to have the packet fragmented. This might actually decrease performance on those specific machines. The rest should all run fine.
Holy outdated driver, batman!
I looked at the date of the NIC driver on my server, and it said 2003. That's weird, because that's a fresh OS load as of last year. Regardless, I updated my driver to the most recent for that NIC (2008) and now my transfers cap around 45MBps, with 1500byte frames!
edit: I tried it with 9k frames and the transfer slows down to 20MBps... odd. Leaving it set to 1500 globally, the transfer starts at 73MBps and is quickly down to about 50MBps, at which point it fluctuates from 46 to 52MBps. That's more like it. I'll play with the 9k jumbo packets another time, when I want to squeeze every last bit of speed out of it. Thanks for your help, guys!
Last edited by ntomsheck : 07-30-2010 at 09:06 PM.
Processors
Motherboards
Chipsets
Memory
Graphics Cards
Storage
Cases and Cooling
Mobile
Systems
Displays
Shows and Expos
The gigabit equipped machines are all running windows. I might consider locating one of my live CDs if the the test returns nothing. Will jperf run from a flash drive in conjunction with a live CD?
Linear Mode
