Which gigabit netgear switch is best for jumboframes ?

I am thinking there is no need to replace the GS108s unless you are seeing some evidence that something needs fixing.

What speeds are you getting now? Are the switches co-located?

StephenB wrote:
I am thinking there is no need to replace the GS108s unless you are seeing some evidence that something needs fixing. What speeds are you getting now? Are the switches co-located?

This is transfer to/from local PC with 6x SSD's (560/535MB/s read/write spec. Corsairs) on Areca 1680IX in raid0 and Intel Pro/1000CT - not very satisfied with the "hacky" upload

It seems unlikely that a switch would behave differently in the upload vs download. All transfers through the switch look pretty much the same (and involve both an upload link and a download link).

Can you bypass the switches and directly connect the two PCs? (manually configuring the IP addresses). That would let you know if the switches and cabling are to blame.

If they are, I'd look at the cabling first, I'd also try different switch ports.

BTW, what performance are you getting with jumbo frames off?

StephenB wrote:

It seems unlikely that a switch would behave differently in the upload vs download. All transfers through the switch look pretty much the same (and involve both an upload link and a download link). Can you bypass the switches and directly connect the two PCs? (manually configuring the IP addresses). That would let you know if the switches and cabling are to blame. If they are, I'd look at the cabling first, I'd also try different switch ports. BTW, what performance are you getting with jumbo frames off?

I will of course get higher transfer rates with direct NAS -> PC transfer !
The whole house are using high quality cat7 ethernet cables - turning off jumboframes in my pc gives severerly lower transferrate with big files - but almost no difference in transfer of small files ?

My "concern" is of course the variable speed in upload - and I can't think of anything else than that it must be the switch/es that causes it
My main computer and the nas are on the same switch - just copying files of different sizes (nas - pc - nas) with a horrible low speed at merely 20Mbps

The other NAS is a another brand with specifications way over netgear readynas ultra6 (and it is putting out much more performance too here home) - and cost's almost twice as much - so I am not complaining on the ultra6 !

EDIT; I think I'll order a Intel Pro/1000MT Dual Port Adapter and connect this pc and the nas directly since the Ultra6 have two nics - that should do it ?

NilsG wrote:
I will of course get higher transfer rates with direct NAS -> PC transfer !

You may be right, though it is not obvious. Even smaller switches are often capable of wire-speed performance - in which case the switch will not degrade performance at all.

NilsG wrote:
The whole house are using high quality cat7 ethernet cables

I believe that "cat 7" is not official. You might have class F or class FA (spec'd for 10 gig ethernet). Maintaining this performance also requires care in wiring up the connectors in the faceplates, and of course any patch cables. So even if the cable itself is good, you might still have end-to-end issues. Does the NAS report any packet loss? (TCP retransmissions, etc).

NilsG wrote:
turning off jumboframes in my pc gives severerly lower transferrate with big files - but almost no difference in transfer of small files ?

That is believable. Jumbo frames generally help because they reduce the interupts per second processing load on the PC. They actually don't improve the network itself. Often the advanced settings on the NIC can be used to improve the non-Jumbo frame performance.

BTW, small file transfers often end up bottlenecked on the NAS. Sequential read/writes are quite fast (writes generally slower with RAID-5 or Raid-6, since the parity blocks also need to be re-written). Copying lots of small files (or updating databases) creates more seeking on the NAS drives, which will slow them down. This is one reason why enterprise storage systems are spec'd in IOs per second, and not MB/s,

NilsG wrote:
My "concern" is of course the variable speed in upload - and I can't think of anything else than that it must be the switch/es that causes it My main computer and the nas are on the same switch - just copying files of different sizes (nas - pc - nas) with a horrible low speed at merely 20Mbps

Trying a direct connect might confirm (or refute) this theory.

Note that most good switches maintain full bandwidth between the ports (e.g, an 8 port gigabit switch would have a 16 gigabit internal bandwidth so it wouldn't slow down when multiple links are active). The connections between your GS108s are of course limited to 1 gbit. So if the switches are co-located you can get higher overall throughput if you consolidate them with a bigger switch. This only matters if you have multiple users loading the links between the switches.

NilsG wrote:
EDIT; I think I'll order a Intel Pro/1000MT Dual Port Adapter and connect this pc and the nas directly since the Ultra6 have two nics - that should do it ?

Just connecting two ethernet cables won't help, unless you also can set up teaming. That requires a managed switch. If you are thinking about dual-homing, then that requires some special care on the network configuration. I strongly recommend that you at least test direct-connect before simply throwing more hardware at the problem.

TX and RX errors is alway zero : Cables are probably the best that is available today

Just now I am replacing disks in this ultra6 nas (some 24 hours restriping left on the third 3TB disk @35% - and three more to come)

As said I have three GS108s - what is meant with co-located ? They are placed in different places in the house since I otherwise must pull a lot of cables to serve this firkin "computerpark" in this house. I also have five XAV5001 powerline adapters (yes I am a huge fan of Netgear products) bang for the buck for home users - but not enterprise products.

I'm at work so i can't see the img, but the hacky upload explains itself with the workload between the smb process and the raid parity process. The nas can't balance the 2 process perfectly, so it first cache the data to write, then slow the upload to process the parity.

~850 Mbps down and some ~520Mbps up - yeah I know it's not "slow" but as said before a huge difference between up/down

I am now in the process replacing these 2TB 7200 rpm disks with WD red 3 TB's - I fear (hehe) that the red disks will be syrup

Big mistake of me to migrate to WD red think - I will know when they all are inn - and that takes time

NilsG wrote:
TX and RX errors is alway zero

That would tend to rule out any cabling.

NilsG wrote:
Just now I am replacing disks in this ultra6 nas (some 24 hours restriping left on the third 3TB disk @35% - and three more to come)... ~850 Mbps down and some ~520Mbps up - yeah I know it's not "slow" but as said before a huge difference between up/down

Are you testing during the resync? That definitely will slow the NAS down

NilsG wrote:
...the hacky upload explains itself with the workload between the smb process and the raid parity process. The nas can't balance the 2 process perfectly, so it first cache the data to write, then slow the upload to process the parity.

I'm not sure I would say it exactly that way, but there is at least one disk read and two disk writes that need to happen for every write to the raid array. However, unlike read requests, this I/O can be can be queued.

You should make sure that the disk write cache is enabled, and the "disable full system journaling" option is also checked on the system/performance tab in frontview. BTW, if you don't have a UPS protecting the NAS you should get one - especially if you are using these settings.

NilsG wrote:
As said I have three GS108s - what is meant with co-located ?

Co-located means in the same location. Yours are not, so consolidating them into one bigger switch isn't possible (unless you rewire the house).

Are you testing during the resync? That definitely will slow the NAS down

No - I Am NOT testing during resync

Would be nice to see other Ultra6 owners tranferrates ? Any threads about that with pictureproof ?