How about teaming? Is it supported under OI? 

Kind regards, 

The out-side

Op 16 nov. 2012 om 17:57 heeft Brian Hechinger <[email protected]> het 
volgende geschreven:

> On Nov 16, 2012, at 11:33, Jim Klimov <[email protected]> wrote:
> 
>> On 2012-11-16 13:12, Brian Hechinger wrote:
>>> 
>>> 
>>> Have you considered vlan trunking?
>> 
>> As he said, they have an external and an internal network segment.
>> It might be an architectural or even a political/compliance requirement
>> to keep the two network segments physically separate.
> 
> Or it may not. We won't know without asking though. :)
> 
>> Well, in this case they have a segment for storage traffic, which may
>> be assumed to be saturated with bulky IOs, and that would compromise
>> performance for "application data" traffic if done by the same switch
>> hardware. Then again, if the said applications would lag because of
>> slow virtual disk components - there may be reason in trunking both
>> networks and separating data by VLANs.
> 
> You aren't magically increasing the usage of the switches if you implement 
> trunking, however. You'd be splitting each network's load across both. With 
> some care the result should be close to net zero on the load front possibly 
> giving more bandwidth to SAN where it is more than likely needed anyway.
> 
>> However, in general two separate switches allow you to do failover or
>> IPMP, not LACP (there are some models that allow LACP over several
>> interconnected switches, which seem like one switch to the connected
>> server or another device for the purpose of link aggregation; Nortel
>> had this for a while, and Cisco may have it in recent models).
> 
> Yes, the Nexus switches do that with their Virtual Port Channel stuff.
> 
>> But if some means of roundrobining over two separate subnets on two
>> switches is an option for the storage and higher-level apps - this may
>> be quite a reasonable option to boost reliability and possibly level
>> out the performance (switch saturation).
> 
> He said iSCSI therefor he should be able to do MPIO with a round robin policy 
> to achieve this. I believe Linux can do this.
> 
>> The possible performance implications (half-gbE for writes, GbE for
>> reads) were addressed by other posters :)
> 
> Yes, I won't be touching that. :)
> 
>> My 2c,
> 
> There, I've added my half a cent's worth. :)
> 
> -brian
> _______________________________________________
> OpenIndiana-discuss mailing list
> [email protected]
> http://openindiana.org/mailman/listinfo/openindiana-discuss

_______________________________________________
OpenIndiana-discuss mailing list
[email protected]
http://openindiana.org/mailman/listinfo/openindiana-discuss

Reply via email to