@misc{oai:ir.soken.ac.jp:00000694,
 author = {内田, 智久 and ウチダ, トモヒサ and UCHIDA, Tomohisa},
 month = {2016-02-17, 2016-02-17},
 note = {Many high energy physics experiments have been searching a new physics beyond the standard model and it requires more precise detection and analyzing system than the current system. It requires many channels of the detectors, a large data size of an event, and a high rate trigger for a detector system. Since the amount of data handled in the system has increased, the required throughput for a data acquisition (DAQ) system is increased as well.<br />　In order to satisfy the requirements, many DAQ systems adopt a distributed computer system and processes two or more events at the same time. A typical system consists of detector sub-systems, an event builder, event processors, and mass storage devices. The detector subsystem processes signals from detectors and sends the event fragment data to the event builder. The event builder collects these event fragment data from many detector sub-systems and builds completion events from it. The event data are sent to event processor. Finally the data are recorded on mass storage devices. Since an event builder communicates between many distributed detector subsystems, its communication technology is essential. Many recent DAQ systems have adopted network technologies for it, which are called network-based DAQ system. Various networks have been adopted, for example, IEEE 802.3 (Ethernet),  asynchronous transfer mode (ATM), and so on. Especially Ethernet is widely used because it is very cost effective infrastructure. A network such as Ethernet is used with network protocols because a network does not have a mechanism for reliable data delivery.<br />　Since the TCP/IP protocol suite is a standard protocol suite of standard operating systems, for example, Linux, UNIX, and so on, many systems adopt it for reliable data delivery. When we employ the standard reliable protocols such as the TCP/IP protocol suite for an event builder, we encounter a serious problem: event fiagments are heading for the same destination at the same time so that data flows are congested and packets are lost in the network. The packet losses decrease transfer efficiency and induce re-transmissions for reliable data delivery. Generally the mechanics of packet losses by congestion is too complex to predict its behavior and performance. Moreover, the re-transmission mechanism makes more complex situations. In order to solve the problem, there is a method in general networks. That is called a Quality of Service (QoS). There are various methods but the main idea is to assign bandwidths of connections and special network devices are used for assigning bandwidths. There are two assignment methods: one assigns a fix bandwidth known as a constant-bit-rate (CBR) method, the other one dynamically assigns bandwidths known as a variable-bit-rate (VBR) method. If transfer rates of DAQ systems are constant, the CBR method solves the problem. Therefore, it is difficult to achieve a high efficiency with the CBR method. On the other hand, the VBR method requires additional protocols to control network devices for bandwidth assignments. Since the protocol assigns a bandwidth, it is also difficult to achieve a high efficiency because the protocol is not able to quickly control the devices.<br />　When we design a highly efficient network, there is another problem which is packet losses by congestion. The packet losses make difficulties with prediction of transfer performance, and, then, a quantitatively network design. Since an even builder is designed for satisfying requirements of an experiment, this is a serious problem. It is rare case in general networks that many senders transmit to the same receiver at same time. Therefore, general reliable protocols are not designed for this case and we can not find a suitable protocol among standard ones. The other problem of using standard protocols is a heavy workload. If we need a high performance data transfer, we should adopt high performance hardware, for example, a high speed CPU, and so on.<br />　In order to solve these problems, we have developed a new communication network protocol. The main idea of the new protocol is an avoiding congestion with a token passing mechanism. Since senders are controlled by the mechanism and only one sender that has a token is permitted to transmit data to a receiver, packet losses are avoided. By this mechanism, we can quantitatively design the system because its difficulty comes from a complicated mechanism of packet losses by congestion.<br />　We introduced a sliding window mechanism to guarantee for reliable data delivery. The mechanism has a reliable data delivery as well as a data flow control between a sender and a receiver. In the mechanism, data transfer is controlled by an acknowledgement that is used for confirming of data transferred, and a sender transmits multiple packets before waiting for an acknowledgement, so the data are transferred in high efficiency.<br />　We implemented the protocol and constructed systems to measure its performance. Since the protocol has a light workload, we implemented the protocol on a small hardware device to demonstrate that. We constructed various systems with the implementation on Ethernet. Since the protocol requires only packet switching function, we were able to use Ethernet hub and facilely constructed those system.<br />　We employ a polling model in queuing theory for a mathematical model of the protocol. But since the model is mathematically simplified, we modified the equations of the model for our systems.<br />　We measured the systems and analyzed these results with the mathematical model. We found that senders fairly transfer data among senders and the total bandwidth is above 90% from the measured results of transfer data rate variations. Since the calculated average message lengths were good agreement with measured results, we can predict it with given an average transfer data rate of senders and an average message waiting-times can be calculated.<br />　We also measured and analyzed various systems which have different number of senders and network topologies. From the results, we found that these system performances can be calculated and we can quantitatively design a large scale network system. And we found that the protocol was more suitable for DAQ systems than TCP with comparison them.<br />　From these results, the protocol is suitable for the DAQ systems and we can conclude that we can quantitatively design and construct a high performance DAQ system with the new protocol., application/pdf, 総研大甲第848号},
 title = {New Communication Network Protocol for a Data Acquisition System},
 year = {}
}