16 Jun, 2016
1 commit
-
TIPC based clusters are by default set up with full-mesh link
connectivity between all nodes. Those links are expected to provide
a short failure detection time, by default set to 1500 ms. Because
of this, the background load for neighbor monitoring in an N-node
cluster increases with a factor N on each node, while the overall
monitoring traffic through the network infrastructure increases at
a ~(N * (N - 1)) rate. Experience has shown that such clusters don't
scale well beyond ~100 nodes unless we significantly increase failure
discovery tolerance.This commit introduces a framework and an algorithm that drastically
reduces this background load, while basically maintaining the original
failure detection times across the whole cluster. Using this algorithm,
background load will now grow at a rate of ~(2 * sqrt(N)) per node, and
at ~(2 * N * sqrt(N)) in traffic overhead. As an example, each node will
now have to actively monitor 38 neighbors in a 400-node cluster, instead
of as before 399.This "Overlapping Ring Supervision Algorithm" is completely distributed
and employs no centralized or coordinated state. It goes as follows:- Each node makes up a linearly ascending, circular list of all its N
known neighbors, based on their TIPC node identity. This algorithm
must be the same on all nodes.- The node then selects the next M = sqrt(N) - 1 nodes downstream from
itself in the list, and chooses to actively monitor those. This is
called its "local monitoring domain".- It creates a domain record describing the monitoring domain, and
piggy-backs this in the data area of all neighbor monitoring messages
(LINK_PROTOCOL/STATE) leaving that node. This means that all nodes in
the cluster eventually (default within 400 ms) will learn about
its monitoring domain.- Whenever a node discovers a change in its local domain, e.g., a node
has been added or has gone down, it creates and sends out a new
version of its node record to inform all neighbors about the change.- A node receiving a domain record from anybody outside its local domain
matches this against its own list (which may not look the same), and
chooses to not actively monitor those members of the received domain
record that are also present in its own list. Instead, it relies on
indications from the direct monitoring nodes if an indirectly
monitored node has gone up or down. If a node is indicated lost, the
receiving node temporarily activates its own direct monitoring towards
that node in order to confirm, or not, that it is actually gone.- Since each node is actively monitoring sqrt(N) downstream neighbors,
each node is also actively monitored by the same number of upstream
neighbors. This means that all non-direct monitoring nodes normally
will receive sqrt(N) indications that a node is gone.- A major drawback with ring monitoring is how it handles failures that
cause massive network partitionings. If both a lost node and all its
direct monitoring neighbors are inside the lost partition, the nodes in
the remaining partition will never receive indications about the loss.
To overcome this, each node also chooses to actively monitor some
nodes outside its local domain. Those nodes are called remote domain
"heads", and are selected in such a way that no node in the cluster
will be more than two direct monitoring hops away. Because of this,
each node, apart from monitoring the member of its local domain, will
also typically monitor sqrt(N) remote head nodes.- As an optimization, local list status, domain status and domain
records are marked with a generation number. This saves senders from
unnecessarily conveying unaltered domain records, and receivers from
performing unneeded re-adaptations of their node monitoring list, such
as re-assigning domain heads.- As a measure of caution we have added the possibility to disable the
new algorithm through configuration. We do this by keeping a threshold
value for the cluster size; a cluster that grows beyond this value
will switch from full-mesh to ring monitoring, and vice versa when
it shrinks below the value. This means that if the threshold is set to
a value larger than any anticipated cluster size (default size is 32)
the new algorithm is effectively disabled. A patch set for altering the
threshold value and for listing the table contents will follow shortly.- This change is fully backwards compatible.
Acked-by: Ying Xue
Signed-off-by: Jon Maloy
Signed-off-by: David S. Miller
06 Mar, 2015
1 commit
-
The ip/udp bearer can be configured in a point-to-point
mode by specifying both local and remote ip/hostname,
or it can be enabled in multicast mode, where links are
established to all tipc nodes that have joined the same
multicast group. The multicast IP address is generated
based on the TIPC network ID, but can be overridden by
using another multicast address as remote ip.Signed-off-by: Erik Hugne
Reviewed-by: Jon Maloy
Reviewed-by: Ying Xue
Signed-off-by: David S. Miller
10 Feb, 2015
3 commits
-
tipc_snprintf() was heavily utilized by the old netlink API which no
longer exists (now netlink compat).In this patch we swap tipc_snprintf() to the identical scnprintf() in
the only remaining occurrence.Signed-off-by: Richard Alpe
Reviewed-by: Erik Hugne
Reviewed-by: Ying Xue
Reviewed-by: Jon Maloy
Signed-off-by: David S. Miller -
Add TIPC_CMD_NOOP to compat layer and remove the old framework.
All legacy nl commands are now converted to the compat layer in
netlink_compat.c.Signed-off-by: Richard Alpe
Reviewed-by: Erik Hugne
Reviewed-by: Ying Xue
Reviewed-by: Jon Maloy
Signed-off-by: David S. Miller -
The new netlink API is no longer "v2" but rather the standard API and
the legacy API is now "nl compat". We split them into separate
start/stop and put them in different files in order to further
distinguish them.Signed-off-by: Richard Alpe
Reviewed-by: Erik Hugne
Reviewed-by: Ying Xue
Reviewed-by: Jon Maloy
Signed-off-by: David S. Miller
27 Nov, 2014
1 commit
-
The node subscribe infrastructure represents a virtual base class, so
its users, such as struct tipc_port and struct publication, can derive
its implemented functionalities. However, after the removal of struct
tipc_port, struct publication is left as its only single user now. So
defining an abstract infrastructure for one user becomes no longer
reasonable. If corresponding new functions associated with the
infrastructure are moved to name_table.c file, the node subscription
infrastructure can be removed as well.Signed-off-by: Ying Xue
Reviewed-by: Jon Maloy
Signed-off-by: David S. Miller
24 Aug, 2014
2 commits
-
The reference table is now 'socket aware' instead of being generic,
and has in reality become a socket internal table. In order to be
able to minimize the API exposed by the socket layer towards the rest
of the stack, we now move the reference table definitions and functions
into the file socket.c, and rename the functions accordingly.There are no functional changes in this commit.
Signed-off-by: Jon Maloy
Reviewed-by: Erik Hugne
Reviewed-by: Ying Xue
Signed-off-by: David S. Miller -
In this commit, we move the remaining functions in port.c to
socket.c, and give them new names that correspond to their new
location. We then remove the file port.c.There are only cosmetic changes to the moved functions.
Signed-off-by: Jon Maloy
Reviewed-by: Erik Hugne
Reviewed-by: Ying Xue
Signed-off-by: David S. Miller
06 May, 2014
1 commit
-
In the previous commits of this series, we removed all asynchronous
actions which were based on the tasklet handler - "tipc_k_signal()".So the moment has now come when we can completely remove the tasklet
handler infrastructure. That is done with this commit.Signed-off-by: Ying Xue
Reviewed-by: Erik Hugne
Reviewed-by: Jon Maloy
Signed-off-by: David S. Miller
18 Jun, 2013
2 commits
-
TIPC has two internal servers, one providing a subscription
service for topology events, and another providing the
configuration interface. These servers have previously been running
in BH context, accessing the TIPC-port (aka native) API directly.
Apart from these servers, even the TIPC socket implementation is
partially built on this API.As this API may simultaneously be called via different paths and in
different contexts, a complex and costly lock policiy is required
in order to protect TIPC internal resources.To eliminate the need for this complex lock policiy, we introduce
a new, generic service API that uses kernel sockets for message
passing instead of the native API. Once the toplogy and configuration
servers are converted to use this new service, all code pertaining
to the native API can be removed. This entails a significant
reduction in code amount and complexity, and opens up for a complete
rework of the locking policy in TIPC.The new service also solves another problem:
As the current topology server works in BH context, it cannot easily
be blocked when sending of events fails due to congestion. In such
cases events may have to be silently dropped, something that is
unacceptable. Therefore, the new service keeps a dedicated outbound
queue receiving messages from BH context. Once messages are
inserted into this queue, we will immediately schedule a work from a
special workqueue. This way, messages/events from the topology server
are in reality sent in process context, and the server can block
if necessary.Analogously, there is a new workqueue for receiving messages. Once a
notification about an arriving message is received in BH context, we
schedule a work from the receive workqueue to do the job of
receiving the message in process context.As both sending and receive messages are now finished in processes,
subscribed events cannot be dropped any more.As of this commit, this new server infrastructure is built, but
not actually yet called by the existing TIPC code, but since the
conversion changes required in order to use it are significant,
the addition is kept here as a separate commit.Signed-off-by: Ying Xue
Signed-off-by: Jon Maloy
Signed-off-by: Paul Gortmaker
Signed-off-by: David S. Miller -
As per feedback from the netdev community, we change the buffer
overflow protection algorithm in receiving sockets so that it
always respects the nominal upper limit set in sk_rcvbuf.Instead of scaling up from a small sk_rcvbuf value, which leads to
violation of the configured sk_rcvbuf limit, we now calculate the
weighted per-message limit by scaling down from a much bigger value,
still in the same field, according to the importance priority of the
received message.To allow for administrative tunability of the socket receive buffer
size, we create a tipc_rmem sysctl variable to allow the user to
configure an even bigger value via sysctl command. It is a size of
three (min/default/max) to be consistent with things like tcp_rmem.By default, the value initialized in tipc_rmem[1] is equal to the
receive socket size needed by a TIPC_CRITICAL_IMPORTANCE message.
This value is also set as the default value of sk_rcvbuf.Originally-by: Jon Maloy
Cc: Neil Horman
Cc: Jon Maloy
[Ying: added sysctl variation to Jon's original patch]
Signed-off-by: Ying Xue
[PG: don't compile sysctl.c if not config'd; add Documentation]
Signed-off-by: Paul Gortmaker
Signed-off-by: David S. Miller
18 Apr, 2013
1 commit
-
Add InfiniBand media type based on the ethernet media type.
The only real difference is that in case of InfiniBand, we need the entire
20 bytes of space reserved for media addresses, so the TIPC media type ID is
not explicitly stored in the packet payload.Sample output of tipc-config:
# tipc-config -v -addr -netid -nt=all -p -m -b -n -ls
node address:
current network id: 4711
Type Lower Upper Port Identity Publication Scope
0 167776257 167776257 1855512578 cluster
167776260 167776260 1216454658 zone
1 1 1 1216479236 node
Ports:
1216479235: bound to {1,1}
1216454657: bound to {0,167776260}
Media:
eth
ib
Bearers:
ib:ib0
Nodes known:
: up
Link
Window:20 packets
RX packets:0 fragments:0/0 bundles:0/0
TX packets:0 fragments:0/0 bundles:0/0
RX naks:0 defs:0 dups:0
TX naks:0 acks:0 dups:0
Congestion bearer:0 link:0 Send queue max:0 avg:0Link
ACTIVE MTU:2044 Priority:10 Tolerance:1500 ms Window:50 packets
RX packets:80 fragments:0/0 bundles:0/0
TX packets:40 fragments:0/0 bundles:0/0
TX profile sample:22 packets average:54 octets
0-64:100% -256:0% -1024:0% -4096:0% -16384:0% -32768:0% -66000:0%
RX states:410 probes:213 naks:0 defs:0 dups:0
TX states:410 probes:197 naks:0 acks:0 dups:0
Congestion bearer:0 link:0 Send queue max:1 avg:0Signed-off-by: Patrick McHardy
Signed-off-by: David S. Miller
01 May, 2012
1 commit
-
Some of the comment blocks are floating in limbo between two
functions, or between blocks of code. Delete the extra line
feeds between any comment and its associated following block
of code, to be consistent with the majority of the rest of
the kernel. Also delete trailing newlines at EOF and fix
a couple trivial typos in existing comments.This is a 100% cosmetic change with no runtime impact. We get
rid of over 500 lines of non-code, and being blank line deletes,
they won't even show up as noise in git blame.Signed-off-by: Paul Gortmaker
02 Jan, 2011
4 commits
-
As the first step in removing obsolete debugging code from TIPC the
files that implement TIPC's non-debug-related log buffer subsystem
are renamed to better reflect their true nature.Signed-off-by: Allan Stephens
Signed-off-by: Paul Gortmaker
Signed-off-by: David S. Miller -
Eliminates routines, data structures, and files that make up TIPC's
user registry. The user registry is no longer needed since the native
API routines that utilized it no longer exist and there are no longer
any internal TIPC services that use it.Signed-off-by: Allan Stephens
Signed-off-by: Paul Gortmaker
Signed-off-by: David S. Miller -
Eliminates routines, data structures, and files that were intended
to allow TIPC to support a network containing multiple clusters.
Currently, TIPC supports only networks consisting of a single cluster
within a single zone, so this code is unnecessary.Signed-off-by: Allan Stephens
Signed-off-by: Paul Gortmaker
Signed-off-by: David S. Miller -
Eliminates routines, data structures, and files that were intended
to allows TIPC to support a network containing multiple zones.
Currently, TIPC supports only networks consisting of a single cluster
within a single zone, so this code is unnecessary.Signed-off-by: Allan Stephens
Signed-off-by: Paul Gortmaker
Signed-off-by: David S. Miller
