Hi Experts.

 

Can anybody help me in this matter???

Already lost 3 days on this subject:

 

I have a Supermicro 6015B-T server with two gigabit eth ports

Each port is connected to a separate network.

In each network is an NAS Array EMC IX12-300R with 12x2TB HGST Drives in it.

 

Both arrays have one share which is NFS enabled and mounted under unRaid.

 

Now i have to copy about 5TB of data in large files:

System makes te following crap:

 

 

Why the copying is not going like in "B" area, but rather like in "A" area -> read from one interface than write to the second one...

And what is the delay in the "C" area.

 

This whole situation drops the copying speed to average 20-30  MB/s, wich is totally inacceptable whereas both of the EMC NAS Arrays can go well beyond 100 MB/s ++

And Yes -> cabling is OK (Cat6A), gigabit routers, structural cabling for 700Mhz, so no issue there.

When I use windows to copy between this shares i got about 50MB/s with computer with one network card and routed trafic on the main router.

 

Attached the diagnostic file as well.

Can anyone help me please ?

 

BR

Chris.

tower-diagnostics-20170428-2007.zip

Edited April 28, 20179 yr by art-informa.pl
typo

Have you tried SMB instead of NFS ?

 

12 hours ago, bonienl said:

Have you tried SMB instead of NFS ?

 

yes

no differene at all...

I don't understand the intended traffic flow. Are you seeing these numbers transferring TO the unRaid server or THROUGH the unRaid server?

 

You said you got 50MB/s using a windows computer. Was that copying from one EMC to the other/unRaid via windows? 

10 hours ago, 1812 said:

I don't understand the intended traffic flow. Are you seeing these numbers transferring TO the unRaid server or THROUGH the unRaid server?

 

You said you got 50MB/s using a windows computer. Was that copying from one EMC to the other/unRaid via windows? 

 

I both cases it is the same configuration with the "copier" in the middle" - so the copying is THROUGH

Let me graph it

 

 

 

If in place of XXXXXXXXXXXXXXXXXXXXXXXXXX is the unraid server (as on the picture) the transfer rate is at top 30 MB/s

If in place of XXXXXXXXXXXXXXXXXXXXXXXXXX is a windows machine, the transfer rate is 50 MB/s

 

Altrough i would expect it to be near full 1GBit....

 

Am i doing something wrong ?

 

When copying files from EMC to unRaid (or vice-versa) it is much better (80MB/s)

 

Am I missunderstanding some basic principles ?

Edited April 29, 20179 yr by art-informa.pl
spelling

50MB(megabytes) = 400Mb(megabits)

Typical HDD speed 300+Mbps

 

You can get higher speed if the files are splitted in two hdd. But that still depends on hdd controller speed. In unraid, if you write to unraid, you would write to 1 hdd, the cache drive. so the write speed is as fast as the cache drive. If you dont have cache drive, the speed would be much much slower.

 

There are also other things at play(network device actual performance, hdd controller, mbb, etc). But, generally, I would say what you experience is normally good. You maxxed your hdd practical speed.

 

If you want to truly measure your unraid transfer rate performance, eliminate as much network device as possible. Test reading from unraid and then test writing on unraid. make sure mover is not active.

49 minutes ago, publicENEMY said:

50MB(megabytes) = 400Mb(megabits)

Typical HDD speed 300+Mbps

 

 

Source and destination are hardware raid arrays capable of 200 MB/s each. So the source and destination is not typical HDD.

Quite strange to me doh...

 

50 minutes ago, publicENEMY said:

 

If you want to truly measure your unraid transfer rate performance, eliminate as much network device as possible. Test reading from unraid and then test writing on unraid. make sure mover is not active.

 

This is not the case.

I want to copy between two hardware arrays.

The performance hardware Array > unRaid or unRaid > hardware Array is about 80MB/s - so no issue there.

 

Also the network devies are not utilized barelly at the moment.

 

I will do another test in next days, connecting two Ethernet ports from the supermicro unRaid server directly one to each of the arrays and see then.

 

I simply cannot understand it why two hardware raid arrays capable of 200MB/s each connected via 4xGb agregations and steered by 2x1Gb copier in between are so piss poor....

I dont think unraid is fully utilizing raid hardware. In fact, the name is even UNRAID. I would think unraid as just a bunch of hard disk with parity data protection. No striping benefits.

 

Why are you using unraid as a router?

3 minutes ago, saarg said:

Why are you using unraid as a router?

 

That was going to be my question based on my understanding of usage.

OK,

 

Let mi clarify some points.

 

1. Generally am not using unRaid as a router. I am using unRaid as unRaid -> no worries. It just haeppened that i needed to copy some data in my network between two independend hardware NAS arrays (EMC IX12-300R). Just haeppened to use unRaid as this machine is "always on".

I though - as normal linux computer - why do not use it for copying it this case???

 

2. My understanding is the following as unRaid is Linux based server -> why he copies data in the weird way. What i would expect in this case, will be eth0 and eth1 working on let's say 800-900 Mbit all the time, till the data is copied from one external array to the other.

Instead of this -> I am obtaining some weird behaviour -> not simultaneous utilization of network interfaces, some dummy delays, etc.

 

3. Regarding the question -> why do I use unRaid as a router -> in fact, in this situation i use it as such, but, copying files between two NAS in generall involves a "computer" inbetween of them. Of course -> i could use some copying scripts on the EMC's but this is not the point.

 

The question may be more general maybe not unRaid specific -> but why, the Linux computer does not utilize both ethernets on full, especially, if the data is not copied TO unRaid drives, no parity calculations should occur, etc...

Seems to be some strange memory buffering in linux system in that case....

 

Thank you all for interesting in this subject

I really look forward to understand what is the problem here.... especially when WIndows do it faster...

 

1. Split the trunk connection between both TP-link switches in 2x2 so VLAN 88 and VLAN 99 traffic runs over a separate physical ports

2. Install the plugin tips-and-tweaks to enable/disable pause frames for your ethernet ports, these affect data throughput behaviour

3. Don't know if possible, but IP forwarding in linux may need to be tweaked

4. I would NOT use this kind of setup in general, but use a proper router instead