> From: ramki Krishnan [mailto:[email protected]] > > > For determining component link imbalance, we are using interface > counters. For tracking the effectiveness of the scheme, we also have > other monitoring information such as 1) Number of times rebalancing was > done and 2) Time since the last rebalancing event. Is there something > else that you think would be helpful? [WEG] yes. You need info about the number of flows being tracked on each interface and their size, possibly also the category (long-lived vs short, large vs small) as well as which flows have been moved and when. You're already building all of this state to provide input to your algorithm, I'm just saying that you need to make it clear in the draft that operators may want to be able to see that data. Like I said, in order to evaluate this as actually being better than the existing hash, to tune the parameters, and to troubleshoot balancing problems, I need to be able to see what it's actually doing, not just the interface counters.
> > >> I realize that in the model you've proposed, we're somewhat limited > because this is using sampled flow data instead of the realtime packet > hash. It may be that this drives a requirement for the granularity of > data being brought into the system in the external mode, and some > requirements about the level of information available via the UI (or > SNMP or XML or whatever) in the automatic hardware-based mode. > > For large flow detection, we propose two modes 1) Sampling Techniques 2) > Automatic Hardware Recognition. The primary difference between the two > modes in terms of performance is the time taken to recognize the long- > lived large flows; there is no loss of accuracy in either mode. In the > automatic hardware based mode for detecting the large flows, it is best > that the load balancing happen in the router. Are you looking for > anything else ? [WEG] Yes, I think maybe I was unclear. I'm looking for diagnostic tools to make this less opaque for an operator. I was noting that it may be difficult to give specific, granular information by flow about the size of each flow and when/if it was moved by the loadbalancing algorithm if you were reliant on sampled flow data, because it wouldn't be complete/granular enough -- it would only show a subset of the flows, not all flows. Thanks Wes This E-mail and any of its attachments may contain Time Warner Cable proprietary information, which is privileged, confidential, or subject to copyright belonging to Time Warner Cable. This E-mail is intended solely for the use of the individual or entity to which it is addressed. If you are not the intended recipient of this E-mail, you are hereby notified that any dissemination, distribution, copying, or action taken in relation to the contents of and attachments to this E-mail is strictly prohibited and may be unlawful. If you have received this E-mail in error, please notify the sender immediately and permanently delete the original and any copy of this E-mail and any printout. _______________________________________________ OPSAWG mailing list [email protected] https://www.ietf.org/mailman/listinfo/opsawg
