Network Storage Protocols Discussions
i have FAS 3020 ONTAP 7.2.3 and cisco 4506 12.2 (20) i ceate multi-mode vif IP Load balancing on netapp and create on cisco etherchannel (no ALCP),
i want to know what is the best performense configuration between netapp and cisco switch? should i change the netapp vif and cisco to LACP ?
See The Solution
I prefer to use LACP over static etherchannel cause the control protocol of LACP uses keep-alives to detect link loss. Just make sure your switch supports LACP.
in other words ,you recomended to move from multi mode to LACP, and get more performence ? what NetApp suggestion ?
LACP and Multi-Mode (better known as: Static Etherchannels) use the same algorithm for determining load-balancing. We are mixing
terms though so lets spend a little time explaining that.
LACP - Stands for Link Aggregation Control Protocol - This is the industry standard for doing etherchannel or port aggregation.
The IEEE designation is 802.3ad
Multi-Mode VIF - Is a NetApp term. There are two types of multi-mode vifs.
a.) Multi-Mode VIF - This is a static etherchannel. This is effectively the standard before IEEE 802.3ad. Basically, networking
manufactures needed a way to interoperate between each other so they arrived at a pre-standard etherchannel. NOTE: The name
static etherchannel, when you run this configuration you are forcing interfaces into a etherchannel statically.
b.) Dynamic Multi-Mode VIF - This is a LACP 802.3ad standards based etherchannel. The unique thing about LACP is that endpoints
exchange PDUs between in each. In these PDUs one device will tell the other about the state of all the links in the channel. These PDUs
continue to be transmitted so that when there is an error one side can alert the other as to the fact that there is a problem.
The next thing to discuss is performance of one versus the other. I mentioned previously that there is no difference in the load-balancing
algorithm, when you create your configurations enabling either LACP or Static Etherchannels you tell the switch and in this case the
storage appliance how you wish to load balance across the channel. You essentially have two choices IP based or MAC based. IP
based is typically preferred reasons why are for another conversation but effectively you take the last octet of the address in the pair and
perform an XOR algorithm then divide that result by the number of active links in the channel. That result equals the link that the
particular source and destination pair will be broadcast on. There is no difference in performance between LACP and Static
Etherchannels thus no difference in Multimode VIFs or Dynamic Multimode VIFs (in NetApp terms).
One of the previous posts mentioned that it is preferred to use LACP over Static Etherchannels because of the ability exchange
information about state. This is precisely correct and is what was referred to above as PDUs. The practical use of this is as follows.
I have seen many customers deploy static etherchannels and have a problem on one side of the link. That problem has something to
do with one of the devices not liking something on the line. The link is still active but one device won't transmit data across one of
the links because of this problem. We have defined a static etherchannel, thus forced the links to transmit across the physical ports.
The only reason a device would stop transmitting is if it had a problem with the physical links (as we just described in our scenario) or
link is lost. Our scenario states that one side has removed the link from the channel but link is still active. This causes the side
that isn't aware of the problem to continue to attempt to load-balance across the physical link. Traffic is essentially sent to a black-hole
and the net effect is users on one floor will work fine and users from another floor will have problems. This is because of the way the
load-balancing algorithm works. Load-Balancing is determined by source and destination pairs one of the devices continues to run
the XOR algorithm on an assumed active link, yet it is not.
LACP solves this problem because of those PDUs. If one side determines there is a problem it lets the other side know. The conversation
goes something like this. Device A transmit PDU informing that I am removing link 1 from channel on my side, Device B receives
mission and removes the same link from channel, thus averting any lost traffic due to black holes.
Hope this helps,
View solution in original post
Not that I can add much to Trey's detailed post, but I think the ultimate answer is....both are fully supported by NetApp with LACP being the better option overall.
thanks you all
Quite welcome. And....don't forget to mark answers as correct or helpful (doesn't necessarily have to be me ).