Mod Multi Packet
(→OpCodes: use the same opcodes as in the old version)
|Line 100:||Line 100:|
Revision as of 12:50, 24 July 2007
Mod Multi Packet Specifications
The mod multi packet is designed as a default container for file Reask related data’s, like ICS or SCT. It can be sent over TCP as well as over UDP. It is in the new mod prot optional but it is strongly recommended to use it to save overhead and have a larger opcode pool available.
The TCP multi Packet(Answer) have the following format:
[hash128 16] // FileHash Entries till data end: [uint8 1] // Sub Opcode [uint16 2] len // entry length [data] // content Next Entry: ...
The TCP packet is sent imminently after the normal eMule Multi Packet, the answer is sent when the mod multi packet is received.
The UDP packet have the exact same format, but it isn't sent together with the file reask ping but as an answer on the "File Reask Ping Acknowledge".
The Extended 4 way file reask ping looks like following:
A B FR_PING ---> // this step may be done over a Kad buddy !!! <--- FR_PING_ACK MOD_MP ---> <--- MOD_MP_ANS
This way was chosen because it allows sending the Mod Multi Packet also to Low ID clients without using the Kad Buddy.
Note: Support for the "Extended 4 way file reask ping" have to be announced separately in the Mod info tag FILE_STATUS_EXTENSIONS
Mod Multi Packet Applications
ICS Extends the regular chunk status with the status of incomplete chunks, it sends a separated status bit field listing chunks that are incomplete but not empty. This is use full for Intelligent Chunk Selection as well as for release related features (v2).
The ICS informations should be accepted in every Multi Packet because all 4 variants should use the same parsing function.
The mod multi packet entry for ICS and other Part based status informations shell have the following format:
[uint16 2] count // PartCount [bitField count/8] // bit encoded status informations
The v1 version sends the informations only in file request answers over TCP, it don't sent them over UDP and it also don't sends them to the source in a file request.
The Neo v2 extension enables sending of the status informations also in file request. The purpose of sending the data in file request is to allow the releasing clients have a better overview witch parts are needed as well as don't apply part hiding for a particular client on parts that he have incomplete.
Whether this informations are to be sent as well over UDP or not is determined by a separated flag (UDP Mod File Status) witch is used as well for other File statuses like SCT or RPS.
The Sub Chunk Transfer feature is mainly only a block status exchange it is sent in file reask answers associated, mod multi packets, not in file requests associated ones, but it can be handled in booth.
The mod multi packet entry for SCT have the following format:
[uint8 1] count // Map Count if count == 0xff [uint16 2] long count // just in case // block maps for selected parts [uint16 2] part // part number [bitField56 7] // [bits 1-53] block status // [bits 54-55] reserved // [bits 56] determines if the map contains verified blocks // next map
Please note that no more than 15 Maps should be sent at once due to the overhead, a recommended limit is min(15,max(5,PartCount/10)). The Maps are sent progressive, only when all have been sent they are resent from the begin and only on TCP, on UDP no resent is done after all maps are out.
Real Part Status is an additional extension of the part status informations, it contains the informations witch parts have been hidden/blocked by releasing client. The Mon Multi packet entry have the exact same format as ICS. It can be sent under 2 different opcodes indicating if the parts are only hidden and are downloadable or if they are completely blocked.
The client must be able to handle the situation when the remote client don't send the extended status informations on every reask(answer) (it must not drop the old status until new status is received).
Clients should only send the extended status over UDP when something in the status changed since the last time.
The extended status should Always be sent over TCP because we don't know if the remote client restarted in the mean time, and lost the old status, if he hadn't he should used a UDP reask. A workaround is also possible: The local client can check if the remote one resent a unchanged status and then resent the own status assuming the remote client had a restart.