We have several HP DL360G6 machines of which some are management servers, some are Hyper-V 2012 cluster, and some are Exchange / File / SQL cluster. The File cluster role is normally running on one node, the Exchange and SQL role on the other. So basically one node of this 2-node cluster is doing file-share only. The issue I have is that on these clusternodes, the shares are slow when writing to. When I copy a large file (let's say 4GB) to one of the local disks (through the admin shares) on one of the other G6 machines, I constantly fill up the GBit network ports, achieving between 110-120MB/sec constantly (partly due to caching of that machine, as it's still writing from memory to disk when the 'copy is done' - taskmanager clearly shows that). The local storage on all machines is just a 2-disk 15krpm RAID1 array. When I do the same to one of the clusternodes it's 'capped' at about 60MB/sec when writing to it - 50% of what I am expecting. Either to the local disks or the cluster disks. Reading from cluster shares or the admin shares of the nodes is not an issue, that gives me 100% network load.
The cluster-storage is located on a HP P2000G3 fibrechannel san, which doesn't have much work to do. The fileshare storage is a RAID6 array, built of 16 450GB 10krpm disks, with VSS enabled. Now don't start about that configuration :) I know RAID6 has a write penalty. However I have benchmarked the storage and it can easyly do 300MB+ on writes, and it easily fills the 4Gb fiber we use when reading. Also when I login to a cluster node, and copy a large file from another machine (through admin share) directly to the clusterdisk, I get the full 120MB/sec again. Disk performace seems no issue here.
On that same storage we have a Hyper-V 2012 cluster of 2 nodes, also built with the very same HP DL360G6 machines with the exact combination. Ofcourse they are running Windows 2012, but before they ran 2008R2 Hyper-V. With either OS we don't have that issue writing to either cluster storage or local storage through network shares on those.
So it seems the 2008R2 failovercluster introduces some throughput cap of sorts. Inititally the chimney offload was enabled, but we have had severe issues with offloading so for now it's disabled. When troubleshooting this issue I re-enabled it on the complete network but it didn't make any difference. I also tried the IPv4 offloading, disabled RSS but that all didn't help. To the clusternodes we added a dualport HP NC112T nic for heartbeat. The client-side network uses the internal NC382i NICs which all other G6 in our network use too. They all have the same firmware and driver level, installed through the HP Support Pack.
So in short we have about 7 G6 machines, of the same hardware configuration, of which the 2008R2 cluster nodes perform quite bad when written to their shares. Are there any caps, limits, preconfigured throttling of any kind known that could cause this?
[edit]
by the way, the cluster validation shows no issues whatsoever.
[/edit]