Hi,
We're currently configuring one of our ESX servers so the network is completly redundant (aim to cover ourselves against NIC, connection or physical switch failure).
We have four LAN connections - two on-board NICs and two on a PCI card. The connections then go to seperate Cisco switches which are connected together.
I have configured two virtual switches - vswitch0, which is has the service console and VMotion, and vSwitch1 which has a port group for virtual machines. Each virtual switch has two pnics (one which is on board and one which is on the PCI card).
I have taken a look at the various recommendations on the forums and ensured that the default origininatiing port ID load balancing is enabled, beaconing is enabled and notify switches is on (on both vswitches). I have then set one active adapter and one standby adapter.
I have tested failover between the two cards (by physicaly pulling cables), and the failover seems to work well - one dropped packet (i'm guessing this is acceptable). When failing back, it isnt quite as seemless - quite a few dropped packets. I have asked the network team to check portfast is enabled on the physical switches - which it is.
Please can someone help? I'm not that clued up on the networking side of things, so if someone could just make some suggestions of things for the network team to take a look at it would be appreciated. I have seen various people mention trunking and etherchannel - but I think this only applies to VLANS? I currently dont have any VLAN tags configured.
A checklist of list of things that need to be done would be nice - something I can pass on to the network guys.
James
Hello,
How many packets do you mean by "few" ?
The scenario you described sounds perfectly normal, I don't think you can prevent this "dropped packet" thing.
The term "trunking" alone is misleading or not precise enought :
\- port trunking (or link aggregation) is the use of multiple ports (thus network cables) to increase the bandwith beyond what a single port is capable
\- VLAN trunking denotes a network link carrying multiple VLANs which are identified by "tags" or "labels"
What ppl are suggesting is you to aggregate some links but not sure how it will work in your configuration since the links you have to bond are connected on different switches.
Sorry, it varies a lot. Minimum ~8, Maximum a few mins worth of pings to the service console from a windows box.
Wow, a few MINUTES ?
Something misconfigured somewhere then...
Do you use spanning tree protocol, or something equivalent ?
Recovery time shouldn't be more than 10 seconds to be acceptable if you ask me, I have the same configuration and recovery time is never above 15 seconds (well...I don't test it that often to be honest :smileysilly:)
Ok, we've done some testing today and it seems that changing the load balancing method to "route based on IP hash" has reducuced our failback times to 1 lost ping - much closer to what I was expecting.
Can anyone envisage any problems with this type of load balancing?
Ok, we've done some testing today and it seems that
changing the load balancing method to "route based on
IP hash" has reducuced our failback times to 1 lost
ping - much closer to what I was expecting.
Can anyone envisage any problems with this type of
load balancing?
No that is what has been suggested to me buy a few people as well...IP Hash seams to work the best
Message was edited by:
CWedge@Amsa
ignore
Message was edited by:
pauliew1978
Hey guys,
Please correct me, but I thought IP hash was only applicable when using etherchannel trunks on Cisco switches, and that this was not possible to do across multiple switches.
Mayur
Thanks for all the replies and suggestions. Seems to be working pretty well now we've got the IP-Hash in place. Will keep an eye on before changing other hosts. Would be nice to have a definitive answer on how it should be setup.
did you get a definative answer on this. How has IP Hasing been performing for you?
