I have heard from a co-worker, that it is possible to transfer more than 1 Gbit/s per stream. The underlaying hardware is no 10 GBit-NICs, it has multiple 1 Gbit Interfaces.
So far he checked my hardware and told me that it is possible, but i should find it out the hard way.
my target is a 2 Gbit/s stream for this:
Terminalserver Win 2008R2 (VM) <-> ESX 5.0 <-> Intel Modular Server <-> HP ProCurve 2810-48G Switch <-> QNAP TS-869U
Currently i still have problems transfering more than 1 Gbit/s with multiple streams, but it seems to be faulty at every place.
VM to VM currently @ 1 Gbit/s
- NIC vmxet3 showing 10Gbit/s
- used no distributed vswitches
- vswitch got configured wit two external interfaces active
- tried here round robin or ip hash - no success
intel modular server with switchign module
- 6 interfaces got configured as static trunk (only option here)
- traffic still passes over one interface, underlaying 3 blades
- switchs ports got here although configured as static trunk
physical Server to another physical server
- both have dual intel server nics, both active, windows server 2008 r2 shows 2 Gbit/s
- configured as 802.3ad dynamic link
- other options (static trunk, adaptive load-balancing, fault tolerance)
- switch ethernet ports have lacp enabled, checked status -> lacp/lag group deteced and active
nas to many hosts, clients
- load only on one interface
- second interface is shown as active
- configured on both sides for 802.3ad
- other options: balance-rr, balance xor, broadcast, balance-tlb, balance.-alb
For my target i think i have to switch at least the nas and servers to alb but its freakin me out, that lacp / static trunk doesn't work at this point. I must have missed something. I'm on this for quite some months, a few hours every week out of business hours. I have read many articles about this and with my unterstanding it should do it. I verify my configuration with iperf resp. jperf or some manual smb transfers from different hosts.
And like always, I'm a little afraid in testing wild configurations with distributed vswitches and hosts in environments that can't afford breakdowns.
I know many informations are still missing, but i try to shorten it at this time. Would help me a lot if we can get this managed.
with kind regards
- Marcel -