Re: Monthly bandwidth does not sum correctly
Posted: Mon Feb 14, 2011 5:10 pm
I found (and fixed) one issue that may be responsible for this problem. Because of the extremely intermittent nature of this issue I can't be sure I got the bug that's been causing all the trouble. I just know I found and fixed a bug that could potentially cause this.
There was a problem in the adjust_ip_for_backwards_time_shift() function of the kernel bandwidth module, so that if there was a backwards shift in the time from one month to the previous one, the old month's bandwidth would get zeroed. There are multiple situations that could potentially result in such a shift, most prominently if there is a mis-match in the time of the router and the NTP time server it is being synced to.
I don't know for sure if this will fix the primary issue. The routine that got fixed is inherently an error correcting routine and I don't know how often that error is showing up. The real problem may be elsewhere. However, I plan on doing a new release with this fix soon and we can see if the error starts showing up again or not.
I'm not confident enough to claim I deserve the bounty yet -- and if this isn't the real problem and someone comes can (1) show that something else is the problem and (2) provide a fix for it... then the bounty is still theirs.
There was a problem in the adjust_ip_for_backwards_time_shift() function of the kernel bandwidth module, so that if there was a backwards shift in the time from one month to the previous one, the old month's bandwidth would get zeroed. There are multiple situations that could potentially result in such a shift, most prominently if there is a mis-match in the time of the router and the NTP time server it is being synced to.
I don't know for sure if this will fix the primary issue. The routine that got fixed is inherently an error correcting routine and I don't know how often that error is showing up. The real problem may be elsewhere. However, I plan on doing a new release with this fix soon and we can see if the error starts showing up again or not.
I'm not confident enough to claim I deserve the bounty yet -- and if this isn't the real problem and someone comes can (1) show that something else is the problem and (2) provide a fix for it... then the bounty is still theirs.
