On Thu, Dec 8, 2011 at 4:55 PM, Erik Zachte ezachte@wikimedia.org wrote:

Asher, "The uncertainty here should be much less than that hanging over all of the stats due to the large but variable random logging packet loss for all wmf traffic that occurred during parts of these months"****

This is actually not the case. We also know exact message loss per squid per hour, inferred from gaps in sequence numbers. ****

So some weeks ago we did patch each hourly file based on weighed average for that hour of losses over all squids. ****

That's great, kudos for that level of attention to detail.