CHANGELOG revision 16a9b66a
12018-03-09 2 3 - [OPTIMIZATION] Merge series of ACKs if possible 4 5 Parsed single-range ACK frames (that is the majority of frames) are 6 saved in the connection and their processing is deferred until the 7 connection is ticked. If several ACKs come in a series between 8 adjacent ticks, we check whether the latest ACK is a strict superset 9 of the saved ACK. If it is, the older ACK is not processed. 10 11 If ACK frames can be merged, they are merged and only one of them is 12 either processed or saved. 13 14 - [OPTIMIZATION] Speed up ACK verification by simplifying send history. 15 16 Never generate a gap in the sent packet number sequence. This reduces 17 the send history to a single number instead of potentially a series of 18 packet ranges and thereby speeds up ACK verification. 19 20 By default, detecting a gap in the send history is not fatal: only a 21 single warning is generated per connection. The connection can continue 22 to operate even if the ACK verification code is not able to detect some 23 inconsistencies. 24 25 - [OPTIMIZATION] Rearrange the lsquic_send_ctl struct 26 27 The first part of struct lsquic_send_ctl now consists of members that 28 are used in lsquic_send_ctl_got_ack() (in the absense of packet loss, 29 which is the normal case). To speed up reads and writes, we no longer 30 try to save space by using 8- and 16-bit integers. Use regular integer 31 width for everything. 32 33 - [OPTIMIZATION] Cache size of sent packet. 34 35 - [OPTIMIZATION] Keep track of the largest ACKed in packet_out 36 37 Instead of parsing our own ACK frames when packet has been acked, 38 use the value saved in the packet_out structure when the ACK frame 39 was generated. 40 41 - [OPTIMIZATION] Take RTT sampling conditional out of ACK loop 42 43 - [OPTIMIZATION] ACK processing: only call clock_gettime() if needed 44 45 - [OPTIMIZATION] Several code-level optimizations to ACK processing. 46 47 - Fix: http_client: fix -I flag; switch assert() to abort() 48 492018-02-26 50 - [API Change] lsquic_engine_connect() returns pointer to the connection 51 object. 52 - [API Change] Add lsquic_conn_get_engine() to get engine object from 53 connection object. 54 - [API Change] Add lsquic_conn_status() to query connection status. 55 - [API Change] Add add lsquic_conn_set_ctx(). 56 - [API Change] Add new timestamp format, e.g. 2017-03-21 13:43:46.671345 57 - [OPTIMIZATION] Process handshake STREAM frames as soon as packet 58 arrives. 59 - [OPTIMIZATION] Do not compile expensive send controller sanity check 60 by default. 61 - [OPTIMIZATION] Add fast path to gquic_be_gen_reg_pkt_header. 62 - [OPTIMIZATION] Only make squeeze function call if necessary. 63 - [OPTIMIZATION] Speed up Q039 ACK frame parsing. 64 - [OPTIMIZATION] Fit most used elements of packet_out into first 64 bytes. 65 - [OPTIMIZATION] Keep track of scheduled bytes instead of calculating. 66 - [OPTIMIZATION] Prefetch next unacked packet when processing ACK. 67 - [OPTIMIZATION] Leverage fact that ACK ranges and unacked list are. 68 ordered. 69 - [OPTIMIZATION] Reduce function pointer use for STREAM frame generation 70 - Fix: reset incoming streams that arrive after we send GOAWAY. 71 - Fix: delay client on_new_conn() call until connection is fully set up. 72 - Fixes to buffered packets logic: splitting, STREAM frame elision. 73 - Fix: do not dispatch on_write callback if no packets are available. 74 - Fix WINDOW_UPDATE send and resend logic. 75 - Fix STREAM frame extension code. 76 - Fix: Drop unflushed data when stream is reset. 77 - Switch to tracking CWND using bytes rather than packets. 78 - Fix TCP friendly adjustment in cubic. 79 - Fix: do not generate invalid STOP_WAITING frames during high packet 80 loss. 81 - Pacer fixes. 82 832017-12-18 84 85 - Fix: better follow cubic curve after idle period 86 - Fix: add missing parts to outgoing packet splitting code 87 - Fix: compilation using gcc 4.8.4 88 892017-10-31 90 91 - Add APIs.txt -- describes LSQUIC APIs 92 932017-10-31 94 95 - [API Change] Sendfile-like functionality is gone. The stream no 96 longer opens files and deals with file descriptors. (Among other 97 things, this makes the code more portable.) Three writing functions 98 are provided: 99 100 lsquic_stream_write 101 lsquic_stream_writev 102 lsquic_stream_writef (NEW) 103 104 lsquic_stream_writef() is given an abstract reader that has function 105 pointers for size() and read() functions which the user can implement. 106 This is the most flexible way. lsquic_stream_write() and 107 lsquic_stream_writev() are now both implemented as wrappers around 108 lsquic_stream_writef(). 109 110 - [OPTIMIZATION] When writing to stream, be it within or without the 111 on_write() callback, place data directly into packet buffer, 112 bypassing auxiliary data structures. This reduces amount of memory 113 required, for the amount of data that can be written is limited 114 by the congestion window. 115 116 To support writes outside the on_write() callback, we keep N 117 outgoing packet buffers per connection which can be written to 118 by any stream. One half of these are reserved for the highest 119 priority stream(s), the other half for all other streams. This way, 120 low-priority streams cannot write instead of high-priority streams 121 and, on the other hand, low-priority streams get a chance to send 122 their packets out. 123 124 The algorithm is as follows: 125 126 - When user writes to stream outside of the callback: 127 - If this is the highest priority stream, place it onto the 128 reserved N/2 queue or fail. 129 (The actual size of this queue is dynamic -- MAX(N/2, CWND) -- 130 rather than N/2, allowing high-priority streams to write as 131 much as can be sent.) 132 - If the stream is not the highest priority, try to place the 133 data onto the reserved N/2 queue or fail. 134 - When tick occurs *and* more packets can be scheduled: 135 - Transfer packets from the high N/2 queue to the scheduled 136 queue. 137 - If more scheduling is allowed: 138 - Call on_write callbacks for highest-priority streams, 139 placing resulting packets directly onto the scheduled queue. 140 - If more scheduling is allowed: 141 - Transfer packets from the low N/2 queue to the scheduled 142 queue. 143 - If more scheduling is allowed: 144 - Call on_write callbacks for non-highest-priority streams, 145 placing resulting packets directly onto the scheduled queue 146 147 The number N is currently 20, but it could be varied based on 148 resource usage. 149 150 - If stream is created due to incoming headers, make headers readable 151 from on_new. 152 153 - Outgoing packets are no longer marked non-writeable to prevent placing 154 more than one STREAM frame from the same stream into a single packet. 155 This property is maintained via code flow and an explicit check. 156 Packets for stream data are allocated using a special function. 157 158 - STREAM frame elision is cheaper, as we only perform it if a reset 159 stream has outgoing packets referencing it. 160 161 - lsquic_packet_out_t is smaller, as stream_rec elements are now 162 inside a union. 163 1642017-10-12 165 166 - Do not send RST_STREAM when stream is closed for reading 167 - Raise maximum header size from 4K to 64K 168 - Check header name and value lengths against maximum imposed by HPACK 169 - Fix NULL dereference in stream flow controller 170 1712017-10-09 172 173 - Hide handshake implementation behind a set of function pointers 174 - Use monotonically increasing clock 175 - Make sure that retx delay is not larger than the max of 60 seconds 176 1772017-09-29 178 179 - A few fixes to code and README 180 1812017-09-28 182 183 - Add support for Q041; drop support for Q040 184 1852017-09-27 186 187 - Fix CMakeLists.txt: BoringSSL include and lib was mixed up 188 1892017-09-26 190 191 - Add support for Mac OS 192 - Add support for Raspberry Pi 193 - Fix BoringSSL compilation: include <openssl/hmac.h> explicitly 194 1952017-09-22 196 197 - Initial release 198