Promise SanLink2

[David Davidson]

Philipp

Who build your system?!

David

[Chris Duffy]

well, you can do some further tweaking on the
windows side for the Intel 10Gbe nic/card….

Why don’t you try in the device manager, network hardware
area:

jumbo frames 9014

Set Receive Side Scaling (RSS) to enabled

Set Receive Side Scaling (RSS) queues to match the CPU logical core count
i.e. as an example,
On an i7 based computer with hyper-threading enabled, set it to say 8
for the number of cores it has….

increase receive buffers to the maximum of 4096

increase transmit buffers to maximum of 16384

and I hope you have some memory on your windows system 😉

see if anything above helps improve things.

[Mike Coley]

Driver code for the sanlink… How about the Sanlink2 Fiber link… I’m smacking my head against the wall over here with one of these things….

[Chris Duffy]

https://www.promise.com/us/Support/downloadcenter

then use search for sanlink2
then on the left select drivers
then on the right you will see the

Sanlink2 10G SFP+ FCS driver for Mac OS X 10.9 and 10.10

That what you need?

[Patrick Brown]

Hi All.

We are having difficulty in our studio with a new server install. The company who installed it have been unable to troubleshoot it so we have abandoned their support and are looking elsewhere for a new tech team. I am not a system administrator, just but I work here as creative professional and doing all I can to help fix the problem in short term

Since the install, we have major network stalling during file transfers, file corruptions, slow directory browsing, slow network transfers, clients crash when reading or browsing server. All symptoms are intermittent.

The setup is

2 x Promise raids connected to Mac Pro (late 2013) running server via Sanlink2 Fibre channel to Thunderbolt (f2101) device. The Mac is connected to Cisco switch via what I think is an odd aggregation, consisting of 1 x Sanlink2 10g Base T to thunderbolt device, aggregated with the 2 x Ethernet ports built in to the mac pro server.

All clients are 1gb Ethernet connected to the switch.

I have two key questions – 1. What is your opinion of the way that the server / switch connection was aggregated? (Ethernet and sanlink2 together)

2. Is there compatibility issues running two thunderbolt sanlink2 devices on the same MacPro?

[Simon Blackledge]

Your all mac based? Or are these PC’s pulling also ?

Are you 100% sure the servers 10Gig ports are aggregated with the 1Gig ports on the switch ? Thats mad.. surprised it even works.

Your feeding the Mac Pro to the switch at 10Gig but the workstations are all Gig yes?

Have you customised the sysctl as in the settings in this thread?

I’d delete all the aggregation. Just connect 1 x 10 GB port from Sanlink to switch and see if that works better to begin with.

Make sure the promise raids and the SanLink are plugged into their own port on the MacPro

You ask can you have 2X sanlink on same machine. Don’t see why not but you only have 1 currently no?

Remember if you connect another machine at 10Gig and start pulling data from one of those Promise raids from the server at say 600MBs and the raid can only do 700MBs then it only leaves 100MBs to everyone else till you stop read/write. I doubt you’d see anything close to 100 even and it will be v choppy.

Delete your aggregation and just start with 1Port feeding the switch/Lan and see where that gets you.

Baby steps.

s

[Patrick Brown]

Hi, Thanks for you response.

We are 80% mac. Our 3D team run PC.

I am 100% sure the 2 server ports were aggregated with the 2 x ethernet cables coming off the sanlink TB to ethernet device, into the switch. The switch seems to be 1Gb only (SGE2010P 48-Port Gigabit Switch)

All workstations are gigabit

Have not customised the sysctl as in the settings in this thread – not 100% sure what this isor where to tweak the settings, but we have a new tech support solution lined up so I will show them the settings you refer too.

Each Sanlink device was plugged into their own thunderbolt port, however I was told they are on the same bus.

As per your suggestion, we removed the aggregation with the Sanlink TB – Ethernet box, and connected with just 2 x built in ethernet ports to the switch. This setup appears more stable, for 1 day so far anyway. I will continue to test this for another few days. Far less reports of network slow down, no corruptions as yet. I will report back if this helps.

Yes – we did have 2 x sanlink devices. The reason I asked about having 2 x sanlink on same machine would cause problems was this – when we disconnected the sanlink2 10 Gb Base-T, i tried plugging the device back into the server later that day. The device was not recognised by OS X or the sanlink utility this time. I wondered if there is a conflict using both a Sanlink2 Fibre Channel (for the raids) on the same machine as a Sanlink2 10 Gb Base-T. Or Maybe this is a sign the Sanlink2 10 Gb Base-T is faulty?

We plan to keep all workstations / clients at 1Gg for now. No 10Gig clients.

Thank you so much for your help and support.

[Chris Duffy]

Just some comments…….

Like the previous guy said 🙂
do not mix 1Gbit and 10Gbit ports in a lagg/bond,
just use one 10Gbe to switch as it can handle plenty of
1Gbit clients.

Second, it is key that the switch you are using
supports symmetric flow-control (flow control
in both directions) i.e. 802.3x standard
on ALL of it’s ports.
A lot of the Cisco’s only support flow-control in one
direction and you will be unhappy since the switch will
be dropping packets when attempting to transfer data
to/from a 10Gbit port and 1Gbit ports….. If the
switch you have does not support flow-control properly
then just get a different switch…..most of the newer
switches have 12/24 gig ports and at least 1-4 10Gbe uplink
ports and they are cheap and work 🙂

Once you stop using the bond, get onto a switch that supports
flow-control properly, then you can take the next step and
enable jumbo frames (9000 bytes as an example) on the Mac
network ports, switch ports and the server 10Gbe port. This
will get you more performance……in addition to what the previous
gentleman stated about getting a good /etc/sysctl.conf file on the
server which will “tune” the network stack fort he 10Gbe port
on this server.

One last thing….Thunderbolt is nice but in some cases where the
Thunderbolt is shared by multiple daisy-chained devices
you have to remember that all the devices are taking/sharing resources
and you will not get the bandwidth that you think you ought
to get…..

It will all work out if you follow the steps offered to
you by others and the above.

[Baule Alessandro]

Hello,

The mechanics is:

Lagg Ethernet – only with same Ethernet velocity, prefer same chipset in the board, (2 x 1 gbits or 2 x 10gb), the mix of the link velocity the system will be based for the link more slow, how all the system work for the more slow, this is valued for memories ram mixed, hds mixed raid, etc.

Files corrupt, look the ram memory’s and the healthy Hds, transfers about network don’t corrupt files in the server.

My system work about the freenas server, with 2x 10gbits Ethernet, one 12 ports 10 gbits switch Netgear, and all HDS in the Freenas are Hibrids HDs Seagate 4TB.

In network mechanicals, all macs thunderbirds Ethernet adapters are connected in the switch 10gbits, and one port is uplink in the other switch 1 Gbits where are connected all 1 gbits machines, the lagg 10gbits are connected directly in the 10gbits switch together with macs.

The system are done for 4k edition simultaneously, mixed with other stations prores 422 edition and other machine making upload of the media files for workflow.

The videos editions are Adobe Premiere and FinalCut, the PCs work with After Effects and others effects programs, and all this working simultaneously .

Remember, for Ethernet 10gbits use 6A cables categorized is fundamentals.

[Baule Alessandro]

Do you are to be use a thunderbolt to Ethernet 2 ports 10gbe to connect a 1 gbits switch?

If yes, because do you don’t use other Ethernet card pcie?

For 1 gbits Ethernet don’t necessit sysctl config.