diff options
author | Luca Muscariello <muscariello@ieee.org> | 2021-04-15 09:05:46 +0200 |
---|---|---|
committer | Mauro Sardara <msardara@cisco.com> | 2021-04-15 16:36:16 +0200 |
commit | e92e9e839ca2cf42b56322b2489ccc0d8bf767af (patch) | |
tree | 9f1647c83a87fbf982ae329e800af25dbfb226b5 /libtransport/src/protocols/prod_protocol_rtc.h | |
parent | 3e541d7c947cc2f9db145f26c9274efd29a6fb56 (diff) |
[HICN-690] Transport Library Major Refactory
The current patch provides a major refactory of the transportlibrary.
A summary of the different components that underwent major modifications is
reported below.
- Transport protocol updates
The hierarchy of classes has been optimized to have common transport services
across different transport protocols. This can allow to customize a transport
protocol with new features.
- A new real-time communication protocol
The RTC protocol has been optimized in terms of algorithms to reduce
consumer-producer synchronization latency.
- A novel socket API
The API has been reworked to be easier to consumer but also to have a more
efficient integration in L4 proxies.
- Several performance improvements
A large number of performance improvements have been included in
particular to make the entire stack zero-copy and optimize cache miss.
- New memory buffer framework
Memory management has been reworked entirely to provide a more efficient infra
with a richer API. Buffers are now allocated in blocks and a single buffer
holds the memory for (1) the shared_ptr control block, (2) the metadata of the
packet (e.g. name, pointer to other buffers if buffer is chained and relevant
offsets), and (3) the packet itself, as it is sent/received over the network.
- A new slab allocator
Dynamic memory allocation is now managed by a novel slab allocator that is
optimised for packet processing and connection management. Memory is organized
in pools of blocks all of the same size which are used during the processing of
outgoing/incoming packets. When a memory block Is allocated is always taken
from a global pool and when it is deallocated is returned to the pool, thus
avoiding the cost of any heap allocation in the data path.
- New transport connectors
Consumer and producer end-points can communication either using an hicn packet
forwarder or with direct connector based on shared memories or sockets.
The usage of transport connectors typically for unit and funcitonal
testing but may have additional usage.
- Support for FEC/ECC for transport services
FEC/ECC via reed solomon is supported by default and made available to
transport services as a modular component. Reed solomon block codes is a
default FEC model that can be replaced in a modular way by many other
codes including RLNC not avaiable in this distribution.
The current FEC framework support variable size padding and efficiently
makes use of the infra memory buffers to avoid additiona copies.
- Secure transport framework for signature computation and verification
Crypto support is nativelty used in hICN for integrity and authenticity.
Novel support that includes RTC has been implemented and made modular
and reusable acrosso different transport protocols.
- TLS - Transport layer security over hicn
Point to point confidentiality is provided by integrating TLS on top of
hICN reliable and non-reliable transport. The integration is common and
makes a different use of the TLS record.
- MLS - Messaging layer security over hicn
MLS integration on top of hICN is made by using the MLSPP implemetation
open sourced by Cisco. We have included instrumentation tools to deploy
performance and functional tests of groups of end-points.
- Android support
The overall code has been heavily tested in Android environments and
has received heavy lifting to better run natively in recent Android OS.
Co-authored-by: Mauro Sardara <msardara@cisco.com>
Co-authored-by: Michele Papalini <micpapal@cisco.com>
Co-authored-by: Olivier Roques <oroques+fdio@cisco.com>
Co-authored-by: Giulio Grassi <gigrassi@cisco.com>
Change-Id: If477ba2fa686e6f47bdf96307ac60938766aef69
Signed-off-by: Luca Muscariello <muscariello@ieee.org>
Diffstat (limited to 'libtransport/src/protocols/prod_protocol_rtc.h')
-rw-r--r-- | libtransport/src/protocols/prod_protocol_rtc.h | 127 |
1 files changed, 127 insertions, 0 deletions
diff --git a/libtransport/src/protocols/prod_protocol_rtc.h b/libtransport/src/protocols/prod_protocol_rtc.h new file mode 100644 index 000000000..f3584f74a --- /dev/null +++ b/libtransport/src/protocols/prod_protocol_rtc.h @@ -0,0 +1,127 @@ +/* + * Copyright (c) 2017-2019 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#pragma once + +#include <hicn/transport/core/name.h> +#include <protocols/production_protocol.h> + +#include <atomic> +#include <map> +#include <mutex> + +namespace transport { +namespace protocol { + +class RTCProductionProtocol : public ProductionProtocol { + public: + RTCProductionProtocol(implementation::ProducerSocket *icn_socket); + ~RTCProductionProtocol() override; + + using ProductionProtocol::start; + using ProductionProtocol::stop; + + void produce(ContentObject &content_object) override; + uint32_t produceStream(const Name &content_name, + std::unique_ptr<utils::MemBuf> &&buffer, + bool is_last = true, + uint32_t start_offset = 0) override; + uint32_t produceStream(const Name &content_name, const uint8_t *buffer, + size_t buffer_size, bool is_last = true, + uint32_t start_offset = 0) override; + uint32_t produceDatagram(const Name &content_name, + std::unique_ptr<utils::MemBuf> &&buffer) override; + uint32_t produceDatagram(const Name &content_name, const uint8_t *buffer, + size_t buffer_size) override { + return produceDatagram(content_name, utils::MemBuf::wrapBuffer( + buffer, buffer_size, buffer_size)); + } + + void registerNamespaceWithNetwork(const Prefix &producer_namespace) override; + + void setConsumerInSyncCallback( + interface::ProducerInterestCallback &&callback) { + on_consumer_in_sync_ = std::move(callback); + } + + private: + // packet handlers + void onInterest(Interest &interest) override; + void onError(std::error_code ec) override; + void produceInternal(std::shared_ptr<ContentObject> &&content_object, + const Name &content_name); + void sendNack(uint32_t sequence); + + // stats + void updateStats(); + void scheduleRoundTimer(); + + // pending intersts functions + void addToInterestQueue(uint32_t interest_seg, uint64_t expiration); + void sendNacksForPendingInterests(); + void removeFromInterestQueue(uint32_t interest_seg); + void scheduleQueueTimer(uint64_t wait); + void interestQueueTimer(); + + core::Name flow_name_; + + uint32_t current_seg_; // seq id of the next packet produced + uint32_t prod_label_; // path lable of the producer + uint16_t header_size_; // hicn header size + + uint32_t produced_bytes_; // bytes produced in the last round + uint32_t produced_packets_; // packet produed in the last round + + uint32_t max_packet_production_; // never exceed this number of packets + // without update stats + + uint32_t bytes_production_rate_; // bytes per sec + uint32_t packets_production_rate_; // pps + + std::unique_ptr<asio::steady_timer> round_timer_; + uint64_t last_round_; + + // delayed nacks are used by the producer to avoid to send too + // many nacks we the producer rate is 0. however, if the producer moves + // from a production rate higher than 0 to 0 the first round the dealyed + // should be avoided in order to notify the consumer as fast as possible + // of the new rate. + bool allow_delayed_nacks_; + + // queue for the received interests + // this map maps the expiration time of an interest to + // its sequence number. the map is sorted by timeouts + // the same timeout may be used for multiple sequence numbers + // but for each sequence number we store only the smallest + // expiry time. In this way the mapping from seqs_map_ to + // timers_map_ is unique + std::multimap<uint64_t, uint32_t> timers_map_; + + // this map does the opposite, this map is not ordered + std::unordered_map<uint32_t, uint64_t> seqs_map_; + bool queue_timer_on_; + std::unique_ptr<asio::steady_timer> interests_queue_timer_; + + // this callback is called when the remote consumer is in sync with high + // probability. it is called only the first time that the switch happen. + // XXX this makes sense only in P2P mode, while in standard mode is + // impossible to know the state of the consumers so it should not be used. + bool consumer_in_sync_; + interface::ProducerInterestCallback on_consumer_in_sync_; +}; + +} // namespace protocol + +} // end namespace transport |