Yesterday, Aug 21st for the second day in a row Bido had problems resulting in their auctions being cancelled until sometime in the future, could be forever we suppose. Since we will be having our own auctions soon this is of great interest to us. We can and will do a little bit of speculation about the reasons and also a little bit of summarizing Bido’s history, as we see it.

From what we’ve read so far there was a problem with servers failing - perhaps because people watching or participating in the auction were refreshing their pages when the data was being pushed to them. Some advice has been to use a server cluster to be more scalable. Underlying is the idea of it perhaps being user behavior that caused the server overload and that this behavior needs to be prepared for by for example a cluster of servers. SQL query caching has also been a suggestion to easy strain on resources.

We can certainly see that it would be difficult to change user behavior, those users having been well conditioned by Ebay et al to refresh the browser page to see new bids. We do it, you have to. Use of clustered servers is a good idea as is making everything as efficient as possible by optimizing SQL queries and the caching of them and the point of this post is to examine things and perhaps use Bido’s experience to learn a lesson without actually experiencing the failure firsthand. Equally important is to not let ourself be drawn into making changes that are not necessary because our situation is different.

Assuming that user behavior was the underlying problem leads to wondering how many users have to misbehave in order for that to become a problem. The first considerations will have to be how many people will we be expecting and how many people do we consider would make our effort a success? Will we require pre-registration and so be able to know fairly closely how many people to expect and then compensate accordingly? Is it necessary for people to watch the domain auction if they aren’t involved?

We also need to keep in mind that we will be auctioning geo-generic domains and Bido was auctioning domains that were rarely geo-related - they were rarely even what we would term a first class generic. Most of Bido’s auctions featured domains that were second rate domains, in our opinion. We had thought about submitting some names but very early on we began to believe that they were using the auction for their own domains - there were 5 or 6 that we figured were a first class domain names but if we were deciding what domains to auction and nobody was submitting better domains than the ones Bido auctioned we would be actively and obviously begging for decent submissions.

After Bido’s initial auctions our feeling is that the quality of the domain names went down and that there is little chance that the amount of users increased as they went along - we strongly suspect that they decreased. We bid on a domain that Bido was auctioning and examined the bidding on more than a few occasions and there weren’t hundreds of bidders - without research we would estimate that for the most part the number would be less than 10. There may well have been numerous people like ourself who watched and didn’t bid but as far as actual bidders go there were less than 10 or more to the point, no great amount.

Were there dozens of people ready to jump in should the auction look as if the domain would go cheaply? Perhaps, the question is whether or not the auction would have ended with a lower selling price had they not been there? We think they played little or no part in the selling price by being there without bidding. They were probably there to see how much their own domain might get compared to the similar domain being auctioned. We say that based on that we also believe that Bido made little effort to contact end-users of the domains at auction, only domainers really were aware of the auctions. The quality of their auctioned domains was so low as to pretty well make it so only inexperienced domainers would bid since we think that many domains of similar or better quality are available for registration and any experienced domainer knows that.

The initial question becomes whether to accommodate watchers or not and if so to what extent? Certainly they can be considered as separate from bidders which allows us to consider our initial requirements based on bidders only. Now the task becomes one of estimating how many bidders should be allowed for. Our first domain to auction will be “PuebloHomeSales.com” and in Pueblo there are at least 200 entities directly related to home sales in Pueblo and that is not considering that there are some other entities who might be worth contacting, one of the general groups being domainers. For the sake of argument we will assume 250 separate people will be contacted, this number would be highest for domains in the real estate industry and much lower for generics like “plumbers” or even “lawyers”. We will therefore assume that this 250 figure would be the maximum amount of bidders if everyone were interested, they won’t be of course so we will arbitrarily choose 100 as the maximum number of bidders that we could be dealing with and if we get that many we will consider the task of attracting bidders to the auction to have been successfully done.

This is we think more than Bido would have had and yet we feel that 100 users oughtn’t to be a problem, however we can see that 100 users refreshing their page every minute would be a problem. The socket connection that is needed must be re-established each refresh which could indeed overwhelm the system. One choice to be examined is that of somehow being able to prevent or discourage refreshing the page. If that can be accomplished then the problem of user misbehavior vanishes. How to do it?

We think that our live auction with a live auctioneer would provide evidence to the user that they are still connected and there would be confirmation by the sounds of the auctioneer’s voice telling the current bid and perhaps a sign displayed near the auctioneer so that if the auctioneer is speaking about something other than the current bid the user will still be reassured because they can see the evidence of the video stream - if the video or audio freezes then there might be cause to consider refreshing the page but not as long as audio/video flow is coherent.

Watchers can view the auction using a more refresh friendly technology and the potential exists to record the auction and let them see the live auction at a later time of their choosing. We see no reason for people not involved in the bidding to watch live and we don’t believe that the absence of these watchers will affect the sale price.

We estimate that under most circumstances we will be dealing with under 20 bidders. We need to consider the pre-registration of the bidders as well. It would also be nice to have a backup policy especially in situations where an expensive domain is being auctioned - not for the bidders’ protection but for the owner of the domain - not to mention our guarantee or our commissions.