Changelog

New features in 20011231 version

Fixed socklib such as to no longer require the presence of the in_addr_t type (useful for compiling udpcast on older Linux distributions)
Fixed --mcast-addr flag, which was ignored in previous versions
Added new --mcast-all-addr flag, in order to override the default broadcast/multicast address used for finding peers

New features in 20040531 version

Udpcast 20040531 has the following new features:

Bugfix for --min-clients feature (uninitialized variable)

New features in 20040417 version

Udpcast 20040417 has the following new features:

Bootloader based on busybox 1.00 pre 10
Kernel 2.4.29

New features in 20040410 version

Udpcast 20040410 has the following new features:

Bootloader based on busybox 1.00 pre 9
Menu system can now reuse IP address gather by PXE environment, rather than needing to do its own DHCP query

New features in 20040222 version

Udpcast 20040222 has the following new features:

Bootloader based on busybox 1.00 pre 7
2.4.25 Kernel

New features in 20030607 version

Udpcast 20030607 has the following new features:

New improved boot loader, based on busybox
new --min-clients, --min-wait, --max-wait and --nokbd flags for unattended sender operation
Misc bugfixes

New features in 20011211 version

Ported it to SuSE 7.3 bootdisk
Half-duplex mode is now default it turns out that performance impact of using full duplex protocol on a network only capable of half duplex is far worse (10%) than in the inverse situation (1%)
Optimized retransmit timer handling (is now adaptative: a sliding average is computed for the observed response time, and if no response is received within that time, plus 10 percent, a second Reqack is sent out)
Hardwired full-duplex slize size to 112 (ideal setting for 100Mbps half-duplex network)
Point-to-point mode is now automatically selected if only one receiver is present.
Bug fix for FEC mode

New features in 20010930 version

Forward error correction:

Transmit redundant information, to enable receivers to locally reconstruct data which has been lost in transit, without requesting it again from the sender.

The sender calculates error correction blocks, which are a kind of checksum of all data blocks in the slice, and sends those to the receivers too. By using those, the receivers can locally reconstruct missing packets without requesting them again from the server

With this, it is even possible to operate in asynchronous mode, with no need to have a communication channel going back from receiver to sender!

Multicast mode:

Rather than broadcasting the data to all machines on the ethernet, whether interested or not, multicasting allows to send data only those machines which are actually interested. Effectiveness of this depends on the Ethernet switch used:

With a hub, or in Tbase2 networks, all machines still get the data, even if multicast can be filtered out at a lower level (i.e. more cost efficiently)
With a switch supporting IGMP, data is only sent out to those ports which have subscribed to it, easying the load on the rest of the network. Note: Some switches may block multicast data altogether (not good), or still forward it to all ports...

Full duplex data scheduling

The sender can now transmit a new slice of data while it is waiting for the receiver's to acknowledge the previous slice. Old versions of udpcast had to wait for the current slice to be acknowledged before transmitting more data.

Point-to-point mode

Useful for further diminishing the load when only sending data to a single destination. This can also obtain the full 100Mbps on mixed speed networks, which would force the lowest speed (10Mbps) on multicast and/or broadcast.