Short and sweet question. I'm currently achieving an average read rate of 36MB/s over gigabit ethernet. I know there are tons of factors but what are the most likely places I should look to get that read performance up closer to the 100MB/s mark (is this even possible?).

 

I have several 3TB drives all connected to 3GB/s ports. My cache drive is quite slow though by comparison but I don't believe this will affect read speeds overall? All other drives connected to either motherboard 3GB/s ports or connected via SAS card.

 

Thanks

What data drive are you reading from? 5400 (Green) or 7200 (regular)?

 

I get 50-60MB/s reading from Green drives WD20EARS.

 

i get almost exactly 36MB/s no matter which drive i'm reading from... weird i know.

Intel NICs. I never got close to full gigabit speeds until I had Intel NICs on both ends.

 

Broadcom makes good NICs too but they don't have good open source drivers.

 

Realtek NICs just aren't very good.

I'd have to agree that Intel NIC are by far the best. 

 

That said, I have Realtek NIC's in my unRAID server as well as my laptop and desktop machine and have no problems getting consistent 100MB/s transfers to either machine.

I get well over 110MB/s from my array with 3TB WD Reds connected to the motherboard SATA ports.

 

I noticed in your config that you're using the SAS-AOC-SASLP-MV8 card => is it plugged into an x4 slot?  [i assume so, since the motherboard has one available]

 

I don't recall exactly which versions it applied to;  but there were a few of the v5 RCs that had transfer speed issues with the MV8's.    Are you running the latest RC (v12a)?

Werid, I get over 110MB 100% of the time (unless im pulling multiple streams). By any chance have you tried a different cable?

I'm running V5 RC12a.

 

All of my 3TB drives are plugged into the motherboard directly (the 3GB/s ports) and the other 4 drives (including cache) run off the Supermicro SAS card.

 

I suppose an obvious place to start would be moving the parity drive onto the 6GB/s port of the mobo directly, but if I understand life correctly that shouldn't make the slightest bit of a difference to read speeds from the array.

 

I do have an Intel NIC in the server btw (forgot to change my sig accordingly) and get the same 36MB/s speed to my windows box via gigabit ethernet or my Macbook Air over 5ghz wi-fi or thunderbolt ethernet adapter. So in summary, it makes zero difference of platform or connection it seems this 36MB/s speed is server end.

 

Thanks for your info thus far... Further ideas on what to change are obviously welcome!

 

edit: Yes, SAS card is plugged into the 4x slot.

You're correct about the cache drive -- it should make NO difference whether it's on a 3GB or 6GB port.

 

Try a different Ethernet cable => if one of the pairs in the cable you're using has a discontinuity (perhaps from an internal break or a poor connection on the connector) that will significantly slow down the transfers (actually I'd expect it to be even slower than you're seeing).

 

Also, are all your cables Cat-6 ??

 

Get down to basics - try a new cat-6 cable directly connected to your unRAID box from a laptop/PC,  and transfer something that way.  If that's full speed, then keep adding bits back until you find the culprit.  I've seen crappy consumer switches and cheap cables both be the cause of an issue like this.

Get down to basics - try a new cat-6 cable directly connected to your unRAID box from a laptop/PC,  and transfer something that way.  If that's full speed, then keep adding bits back until you find the culprit.  I've seen crappy consumer switches and cheap cables both be the cause of an issue like this.

 

Absolutely no need to have cat 6. Cat 5e is more than adequate.

 

36MB/sec shows that everything is connected via gigabit as the limit for 100 is 12.5MB/sec.

 

 

For reference, my Realtek 8111L was easily able to hit the max gigabit speed of ~110-115MB/sec from any drive that was able to supply it and pretty much every 3.5" > 1TB drive can do that over at-least a portion of the disk.

While Cat-5e is indeed fine, there's such a small difference in cost that if you're buying new cables there's no reason not to simply use Cat-6.

 

As for 36Mb showing that everything is fine -- not necessarily.  It does show the network is Gb;  but it doesn't necessarily mean the cable is good ... as I noted earlier, a break in any pair can dramatically impact the actual transfer speeds -- often in some strange ways.

 

By far the first thing I'd try to see if it makes a difference is to use a new cable from the server to the switch (or directly to a PC).

 

Great tips guys - I will try all the above suggestions before cracking open the case and post my findings.