Wi-Fi is a watershed in the experience of WiFi users on campus. Ask a regular student."How's the school network?"He won't discuss the export bandwidth or certification with you. He'll tell you."The network was cut off when we got to the teaching building, block B."Or..."From the library to the cafeteria, we have to re-log.". The roots of these problems are roaming. But for information centres, walking is not just wireless; if it is not handled properly with the billing system, it can lead directly to misdirection of billing data, repeated deductions and even a user being kicked off the line without any reason.
The billing chain triggered by the roaming is more complicated than expected.
One student moves on campus, and the cell phone moves from AP1 signal to AP2, and much more goes on at the bottom than users see. First, a wireless 802.11r or 11k/v fast-walking protocol completes the APP switch, then the network access layer certifier migration, and finally the billing system level updates.
If the three events are not dealt with in the right time series, there are all sorts of weird questions. The most common ones are: after a cruise, new APs send an Accoutring-Start package to RADIUS and old APs send an Accundating-Stop bag, but two packages arrive in reverse order on the RADIUS server.——The Stop bag arrives first, Start package later. If the billing system does not process the check, it ends as a normal session and uses Start as a new line, creating a blank window for the billing. Users actually have been online, but there is a gap in the billing record where this blank flow goes?
Guarantee of continuity of the RADIUS no-state agreement
The RADICUS protocol is an unstated UDP-borne agreement, designed without considering trans-AP transfers of session status. The standard RADIUS has the Acct-session-Id attribute, which theoretically should be used as a Ssession-Id for one and the same user to go through it all the time, and do not change Session-Id. But the reality is that many manufacturers' APs will recreate a new Sésion-Id after they have travelled around, leading to costing systems thinking that these are two separate sessions.
The key to resolving this problem is that the system itself does a conversational connection on the application level, not relying solely on RADIUS Ssession-Id. This is done by: when it receives Accounting-Stop, the billing system does not immediately close the account ' s billing session but waits for a very short interval (e.g. five seconds) to see if the same user ' s MAC address has arrived at another AP with a new Acoutating-Start. If there is, it suggests that it is a roaming rather than a real downlink, and the costing context should be inherited rather than reset. The waiting window is time-consuming, too long a user experience (the user has been slow to see its own real-time flow statistics after the cruise) and too short for error (in case the user actually breaks off the web and reconnects, there is also a small gap between the two operations).
There is a more complex situation: three layers of roaming. Students wandered across subnets, IP addresses changed, but the MAC address remained unchanged. For a billing system, two IPs appeared for one or both users. This cannot be judged by an MAC address alone. The user account number, MAC address, VLAN, AP location needs to be combined and cross-lined in the billing system, judging by fingerprint matching that the same user is moving.
The most easy pit to step on in the roaming bill: time and double counting of traffic
If the blanks are a loss of money, the double bill is an overcharge, which causes a much larger number of complaints than the former. The pattern that leads to the repetition of fees is usually: old APs who travel before they do send an Accoutating-Stop with cumulative traffic statistics from up-to-road to the time when they do it; and new APs who do so after the cruise start counting from zeros. But if the new AP reports all traffic since it was online when sending the first Accounding-Interim-Update, and the billing system has been cleared once before the cruise, then the deduction is repeated.
The root of the problem is inconsistent in the characterization of AP and billing systems. Some APs report incremental traffic (from previous Interim to present value) each time, while others report cumulative flows (from top-line to current total). If the billing system does not identify, adding cumulative traffic as incremental traffic will repeat the cumulative amount. So the billing system must have the ability to recognize whether Acct-Input-Octets are cumulative or incremental values, usually by: if this is a smaller number than last time, then AP restarts or is a new session and should be treated in incremental rather than cumulative terms.
The storm of a high density.
High-density scenes like school gyms, report halls and canteens, with dozens of devices traveling around between dense APs, a single event."The storm."The storm is not a bandwidth but a RADIFUS service per second. A normal RADIUS server has about thousands of authentication packages per second. But in the trance storm, each device triggers two to four RADIUS bags (Stop, Start, and possibly Interim), multiplied by the number of devices that can travel simultaneously, with peaks reaching tens of thousands per second.
The system of billing to cope with the storm is structured in two ways. First, the UDP buffer at the RADIFAS receiving end should be large enough not to be lost because of the instant amount of bagage.——UDP drops the package in a RADIUS scene amount to a direct loss of billing data. Second, backend processing of the bill engine is to be anathetic, with the front-end RADIUS component only receiving, timetamping and writing messages in queues, and back-end costing modules are being consumed from the queue without blocking reception at the front end.
There is also a means of working from the side of the network: configuration on AC or wireless controller"♪ Walking sticky ♪"But the time of glue cannot be set too long, or the user's arrival to the signal core area of the new AP is stuck on old AP, which leads to a decline in experience. The general suggestion is that one-to-two-minute sticky time threshold triggering roaming.