Bound two NIC in a virtual one to increase the bandwidth

[Bruno Seghers]

Hi,

I'm running OpenVMS 8.3-1h1 on Itanium blades.
I would like to bound 2 Gbit NIC in one virtual interface to have a bandwith of 2 Gbit.
Is there a way to do that ?

Thanks for your help

Best regards

Seghers Bruno
Belgium

[Richard Whalen]

What TCP/IP stack are you using?

MultiNet and TCPware can do it with COMMON_LINK functionality. This allows them to queue packets to be transmitted to the least busy interface. It will also detect a failed interface and avoid using it. Only one interface is used for receives. For common link to work correctly they must feed into a single switch.

Are you really keeping a 1Gb interface busy with transmits, or are you trying to even out the heavy load times? You'll need to plug the interfaces into a high end switch to avoid moving the bottle neck from the system to the switch.

[Heinz W Genhart]

Hi Bruno

If you are using HP TCP/IP Services for OpenVMS, then you can configure Fail Safe IP.
This way you double the bandwidth and you have a save configuration in case a interface fails.

You can find the documentation for Fail Safe IP in chapter 5 of this manual http://h71000.www7.hp.com/doc/83final/6526/BA548_90006.PDF

Hope that solves your problem

regards

Geni

[Bruno Seghers]

Hi,

Thanks for answers.
We are using TCPIP V5.6 ECO2 on OpenVMS 8.3-1H1.

I read faileSafeIP documentation and it's said : "The failSAFE service monitors an interface and takes appropriate action
upon detecting interface failure or recovery. failSAFE IP provides IP address
redundancy by requiring the same IP address to be configured on multiple
interfaces. Only one instance of each IP address is active at any time; the other
duplicate IP addresses are in standby mode."

If only one instance is active at a time, I suppose the bandwidth is not increased.

I dont find something on the net about this.

Thanks for help

Regards

Seghers Bruno
Atos Worldline Belgium

[Wim Van den Wyngaert]

Bruno,

Failsafe IP is failover. Just as define failover device in lancp (but here for all protocols).

Wim

[Hoff]

In some cases, you have that capability automatically, but I'm going to guess you're using IP and specifically TCP/IP Services here.

TCP/IP Services provides the following options:

http://64.223.189.234/node/859

You might be able to get where you need, but AFAIK there is no good (bi-directional) approach toward teaming or ganging or link aggregation using TCP/IP Services. There are some OpenVMS Technical Journal articles on the topic, if you want further reading.

One of the third-party IP stacks might well have some useful capabilities here.

The HP approach here has seemingly been toward 10 GbE connections and 10 GbE NICs.

[Heinz W Genhart]

Hi Bruno

I guess my description was not complete.

A small story (only important things mentioned):

We had a configuration with 2 AS4100 with one DE600 each as boot nodes and 180 Satellites. If this configuration was down, it takes 1.5 hours to boot.
Then we exchanged the AS 4100 with 2 ES45 with 2 DE602 each and MSA1000 for the storage.

We configured Failsafe IP over the 4 nic's. Then, and this point is very important, we defined the 4 addresses of each machine within the dns server.
Example:

node1 172.1.1.1,
172.1.1.2,
172.1.1.3,
172.1.1.4

The dns server changes the order of those addresses with a round-robin algorithm. This means, first connect to node1 uses 172.1.1.1, second connect 172.1.1.2, third connect 172.1.1.3 and so on. We were also using DECnet over Ip. So even the DECnet traffic used all those 4 nic's.
This way we were using all 4 nics and we had a bandwidth of 4 times the nic speed. We could decrease the boot time of our 180 satellites from 1.5 hours to 15 min. This because we were using all the 4 nic's on a much faster machine, with faster storage.
Failsafe ip used this way can increase the total bandwidth of the machine, but you are right, not the bandwidth of a single connection.

The advantage of failsafe ip, if configured well is, that all the 4 addresses will always be available. Even if one nic fails, this address will move to another nic.
If you are using a dns server as described, you can use all the configured nic's all the time.

If you configure lan failover, then you can use only one nic at the time. If one of the nic's within the failover set fails, then you are using another interface. But only one nic at the time.

Maybe this may solve also your problem

Regards

Geni

[Bruno Seghers]

Thanks to all for your answers

Bruno Seghers

[Wim Van den Wyngaert]

Geni,

A satelite boot (without system disk) uses mscp to get access to the system disk on the server. Thus scs/sca. DNS has no effect on this protocol. So I don't understand that DNS could improve the boot.

However, 1 boot requires about 150 MB. Thus 27 GB for 190 stations. That's about the maximum for a 100 MBit card during an hour.
With 4 cards you would end up with 15 minutes (load balncing done by sca).

Wim

[Tom O'Toole]

Itanium satellite boot uses BOOTP I think.