Search

Energy aware server clusters aim to reduce power consumption at maximum while keeping QoS(Quality of Service) compared to energy non-aware server clusters. They adjust the power mode of each server in a fixed or variable time interval to let only the minimum number of servers needed to handle current user requests ON. Previous studies on energy aware server cluster put efforts to reduce power consumption further or to keep QoS, but they do not consider energy efficiency well. In this paper, we propose an energy efficient cluster management based on autonomous learning for energy aware server clusters. Using parameters optimized through autonomous learning, our method adjusts server power mode to achieve maximum performance with respect to power consumption. Our method repeats the following procedure for adjusting the power modes of servers. Firstly, according to the current load and traffic pattern, it classifies current workload pattern type in a predetermined way. Secondly, it searches learning table to check whether learning has been performed for the classified workload pattern type in the past. If yes, it uses the already-stored parameters. Otherwise, it performs learning for the classified workload pattern type to find the best parameters in terms of energy efficiency and stores the optimized parameters. Thirdly, it adjusts server power mode with the parameters. We implemented the proposed method and performed experiments with a cluster of 16 servers using three different kinds of load patterns. Experimental results show that the proposed method is better than the existing methods in terms of energy efficiency: the numbers of good response per unit power consumed in the proposed method are 99.8%, 107.5% and 141.8% of those in the existing static method, 102.0%, 107.0% and 106.8% of those in the existing prediction method for banking load pattern, real load pattern, and virtual load pattern, respectively.

Due to the tremendous amount and its rapid increase of network traffic, the performance of network equipments are becoming an important issue. Network filtering is one of primary functions affecting the performance of the network equipment such as a firewall or a load balancer to process the packet. In this paper, we propose a cache based tri method to improve the performance of the existing tri method of searching for network filtering. When several packets are exchanged at a time between a server and a client, the tri method repeats the same search procedure for network filtering. However, the proposed method can avoid unnecessary repetition of search procedure by exploiting cache so that the performance of network filtering can be improved. We performed network filtering experiments for the existing method and the proposed method. Experimental results showed that the proposed method could process more packets up to 790,000 per second than the existing method. When the size of cache list is 11, the proposed method showed the most outstanding performance improvement (18.08%) with respect to memory usage increase (7.75%).

In the server cluster environment, one of the ways saving energy is to control server`s power according to traffic conditions. This is to determine the ON/OFF state of servers according to energy usage of data center and each server. To do this, we need a way to estimate each server`s energy. In this paper, we use a software-based power consumption estimation model because it is more efficient than the hardware model using power meter in terms of energy and cost. The traditional software-based power consumption estimation model has a drawback in that it doesn`t know well the computing status of servers because it uses only the idle status field of CPU. Therefore it doesn`t estimate consumption power effectively. In this paper, we present a CPU field based power consumption estimation model to estimate more accurate than the two traditional models (CPU/Disk/Memory utilization based power consumption estimation model and CPU idle utilization based power consumption estimation model) by using the various status fields of CPU to get the CPU status of servers and the overall status of system. We performed experiments using 2 PCs and compared the power consumption estimated by the power consumption model (software) with that measured by the power meter (hardware). The experimental results show that the traditional model has about 8-15% average error rate but our proposed model has about 2% average error rate.

A mobile 3-D display processor with a subdivider is presented for higher visual quality on handhelds. By combining a subdivision technique with a 3-D display, the processor can support viewers see realistic smooth surfaces in the air. However, both the subdivision and the 3-D display processes require a high number of memory operations to mobile memory architecture. Therefore, we make efforts to save the bandwidth between the processor and off-chip memory. In the subdivider, we propose a recomputing based depth-first scheme that has much smaller working set than prior works. The proposed scheme achieves about 100:1 bandwidth reduction over the prior subdivision methods. Also the designed 3-D display engine reduces the bandwidth to 27% by reordering the operation sequence of the 3-D display process. This bandwidth saving translates into reductions of off-chip access energy and time. Consequently the overall bandwidth of both the subdivision and the 3-D display processes is affordable to a commercial mobile bus. In addition to saving bandwidth, our work provides enough visual quality and performance. Overall the 3-D display engine achieves 325 fps for 480×320 display resolution.

All the servers in a traditional server cluster environment are kept On. If the request load reaches to the maximum, we exploit its maximum possible performance, otherwise, we exploit only some portion of maximum possible performance so that the efficiency of server power consumption becomes low. We can improve the efficiency of power consumption by controlling power mode of servers according to load situation, that is, by making On only minimum number of servers needed to handle current load while making Off the remaining servers. In the existing power mode control method, they used a static policy to decide server power mode at a fixed time interval so that it cannot adapt well to the dynamically changing load situation. In order to improve the existing method, we propose a dynamic server power control algorithm. In the proposed method, we keep the history of server power consumption and, based on it, predict whether power consumption increases in the near future. Based on this prediction, we dynamically change the time interval to decide server power mode. We performed experiments with a cluster of 30 PCs. Experimental results show that our proposed method keeps the same performance while reducing 29% of power consumption compared to the existing method. In addition, our proposed method allows to increase the average CPU utilization by 66%.

Caching web pages is an important part of web infrastructures. Medium to large-scale infrastructures deploy a cluster of servers to solve the scalability and storage problems inherent in caching. In this paper we present dynamic information-based scalable hashing that evenly hashes client requests to a cluster of cache servers, resulting in performance scalability. Runtime information is used to determine when and how to cache pages. Cached pages are stored and retrieved mutually exclusively to/from all the servers to minimize the use of storage, resulting in storage scalability. We set up an experimental environment consisting of various machines, including client servers, a cluster of 16 cache servers, and a load balancer. We demonstrate through experimental results that dynamic information-based scalable hashing maximizes both performance scalability and storage scalability while the existing approaches do only either one of the two. Copyright © 2011 John Wiley & Sons, Ltd.

Adaptive clustering aims at improving cluster utilization for varying load and traffic patterns. Locality-based least-connection with replication (LBLCR) scheduling that comes with Linux is designed to help improve cluster utilization through adaptive clustering. A key issue with LBLCR, however, is that cluster performance depends much on a single threshold value that is used to determine adaptation. Once set, the threshold remains fixed, regardless of the load and traffic patterns. If a cluster of PCs is to adapt to different traffic patterns for high utilization, a good threshold has to be selected and used dynamically. We present in this paper an adaptive clustering framework that autonomously learns and adapts to different load and traffic patterns at runtime with no administrator intervention. The cluster is configured once and for all. As the patterns change, the cluster automatically expands/contracts to meet the changing demands. At the same time, the patterns are proactively learned so that when similar patterns emerge in the future, the cluster knows what to do to improve utilization. We have implemented this autonomous learning method and compared it with LBLCR using published Web traces. Experimental results indicate that our autonomous learning method produces high performance scalability and adaptability for different patterns. On the other hand LBLCR-based clustering suffers from performance scalability and adaptability for different traffic patterns since it is not designed to obtain good threshold values and use them at runtime.

In this paper, a heterogeneous 3D-media processor is presented, which supports all 3-D display applications by combining a 3-D display IP with a 3-D graphics IP and a stereo video decoder. For mobile environments, adaptive power management scheme is proposed, which saves power consumption up to 186 mW by turning off idle functional blocks based on a target application, a target performance, and the run-time ratio between different IPs. As a result, the minimum power consumption of the processor is only 15 mW, while the overall power consumption is 201 mW. As well as the reduction of power consumption, this work shows impressive performance improvement. The proposed fast modulo operators and adopted division-free algorithm reduces the critical latencies of 3-D display image processing. The proposed fast datapath with parallel architecture increase synthesis rate up to 116 fps which is 17 times faster than a previous work. In addition, reordered operation sequence fixes memory bandwidth regardless of the number of images to be produced. In the 3-D graphics IP and the decoding IP, redundant datapath are merged using an IEEE 754 compliant floating-point vector unit to save both chip area and power consumption, which even reduces the critical latency by 30%.

Chord is a peer-to-peer look up algorithm based on a distributed hash table protocol in wired IP networks, exploiting the advantage of scalability for large-scale of distributed applications. However, deploying Chord into mobile networks should be inherently accompanied with supplementary network traffics to maintain the hash key mapping rules because of a high rate of joining and leaving nodes. This paper proposes an enhanced reactive Chord for mobile networks, which can reduce network traffics and achieve fast lookup services. For this purpose, a conventional Chord is modified to act reactively, and then the table activity checking feature is devised into it, called enhanced reactive Chord. Simulation results show that the proposed Chord can decrease network traffics by an average of 41.3% maintaining same setup latency, compared with conventional Chord in mobile networks.

A DHT(Distributed Hash Table) based P2P is a method to overcome disadvantages of the existing unstructured P2P method. If a DHT algorithm is used, it can do a fast data search and maintain search efficiency independent of the number of peer. The peers in the DHT method send messages periodically to keep the routing table updated. In a mobile environment, the peers in the DHT method should send messages more frequently to keep the routing table updated and reduce the failure of a request. Therefore, this results in increase of network traffic. In our previous research, we proposed a method to reduce the update load of the routing table in the existing Chord by updating it in a reactive way, but the reactive method had a disadvantage to generate more traffic than the existing Chord if the number of requests per second becomes large. In this paper, we propose an adaptive method of routing table update to reduce the network traffic. In the proposed method, we apply different routing table update method according to the number of request message per second. If the number of request message per second is smaller than some threshold, we apply the reactive method. Otherwsie, we apply the existing Chord method. We perform experiments using Chord simulator (I3) made by UC Berkeley. The experimental results show the performance improvement of the proposed method compared to the existing methods.

An SMTP(Simple Mail Transfer Protocol) server was designed for delivering and storing user’s email across a network. An SMTP server can be distributed as multiple servers for large service, huge users or massive emails. An SMTP server can be constructed by multiple servers with separating file system as email storing space, and each SMTP server can usually share each file system by using the NFS(Network File system). However the NFS is originally designed for sharing each file system across a network, and contains all attributes and features of regular file system. Using this NFS for email storing space of SMTP servers, it makes overhead due to unnecessary work of regular file system.In this paper, we propose a method to do directly operation with socket for reducing work overhead caused by the NFS. For doing directly operation with socket, this paper defines information based storing space structure and a protocol for storing emails. We performed experiments using Mailstone made by Netscape. The experimental results show the performance improvement of the proposed method compared to the existing method.Keywords:Email, SMTP Server, Mail Server, Distributed Architecture, NFS(Network File System)

In this paper, a power efficient vertex processor for mobile graphics applications is presented. A four-threaded and four-issue expanded VLIW datapath with a quad-float vertex texture fetcher is proposed by exploiting graphics specific characteristics after evaluation of several candidate architectures. Instruction-level power control methods such as operand sharing and writeback re-allocation along with operand isolations and gated clocks result in 40.4% and 82% reduction in energy dissipation and energy delay product compared to the most widely used single threaded SIMD. The proposed processor with the optimized datapath and vertex caches implemented in a 0.18- mum 1P4M CMOS process achieves 186-Mvertices/s geometry performance which is the best result among the processors that are IEEE-754 compliant.

In this paper, we propose an architecture of tessellation hardware to save memory bandwidth in a mobile multimedia processor. To reduce the implementation overhead, floating-point computations of tessellation are accelerated by the conventional GPU pipeline, and only tessellation-specific control logic is handled by an additional hardware unit. Tightly coupled with a vertex shader, the additional unit dynamically produces topological configurations and parametric coordinates of refinement patterns in the type of indexed triangle strips for object-level adaptive tessellation. The topological configurations improve the efficiency of the vertex cache so as to avoid redundant shader operations. Since the proposed tessellator is area-efficient and does not require intermediate memory accesses, its architecture is especially appropriate for the mobile environment, which adopts a shared bus and unified external memory architecture. The proposed tessellator is fabricated on a chip using 0.18@mm CMOS technology. With 6.2% additional hardware for a dual-core vertex shader, the implemented chip performs 120Mvertices/s vertex shading and saves memory bandwidth up to 250 times in tessellation.

The signature hashing algorithm[9] provides the fast pattern matching speed for network IPS(Intrusion Prevention System) using the hash table. It selects 2 bytes from all signature rules and links to the hash table by the hash value. It has an advantage of performance improvement because it reduces the number of inspecting rules in the pattern matching. However it has a disadvantage of performance drop if the number of rules with the same hash value increases when the number of rules are large and the corelation among rules is strong. In this paper, we propose a method to make all rules distributed evenly to the hash table independent of the number of rules and corelation among rules for overcoming the disadvantage of the signature hashing algorithm. In the proposed method, it checks whether or not there is an already assigned rule linked to the same hash value before a new rule is linked to a hash value in the hash table. If there is no assigned rule, the new rule is linked to the hash value. Otherwise, the proposed method recalculate a hash value to put it in other position. We implemented the proposed method in a PC with a Linux module and performed experiments using Iperf as a network performance measurement tool. The signature hashing method shows performance drop if the number of rules with the same hash value increases when the number of rules are large and the corelation among rules is strong, but the proposed method shows no performance drop independent of the number of rules and corelation among rules.

With increasing multicore system, much effort has been put on the performance improvement of its application. Because multicore system has multiple processing devices in one system, its processing power increases compared to the single core system. However in many cases the advantages of multicore can not be exploited fully because the existing software and hardware were designed to be suitable for single core. When the existing software runs on multicore, its performance improvement is limited by the bottleneck of sharing resources and the inefficient use of cache memory on multicore. Therefore, according as the number of core increases, it doesn`t show performance improvement and shows performance drop in the worst case. In this paper we propose a method of performance improvement of multicore system by applying Flow-Level Parallelism to the existing TCP/IP network application and operating system. The proposed method sets up the execution environment so that each core unit operates independently as much as possible in network application, TCP/IP stack on operating system, device driver, and network interface. Moreover it distributes network traffics to each core unit through L2 switch. The proposed method allows to minimize the sharing of application data, data structure, socket, device driver, and network interface between each core. Also it allows to minimize the competition among cores to take resources and increase the hit ratio of cache. We implemented the proposed methods with 8 core system and performed experiment. Experimental results show that network access speed and bandwidth increase linearly according to the number of core.

Pastry overlay network is one of structured P2Ps using DHT(Distributed Hash Table). To reduce the number of messages among nodes, Rosary and LAR have been proposed by exploiting spatial locality and caching, respectively, in the Pastry. Rosary consists of Inter-Pastry and Intra-Pastry. A root node is assigned as a representative in each Intra-Pastry and it has the responsibility of Inter-Pastry and Intra-Pastry routing. Therefore, Rosary has several disadvantages; 1) low fault tolerance in case of root node failure 2) routing hop count increases because of the use of root nodes compared to the existing structured P2Ps, and 3) the communication load is concentrated in some specific areas. LAR has inefficient problems in that caching is not distributed among nodes in Intra-Pastry and caching is used by only nodes in the Intra-Pastry.In this paper, we propose an improved Pastry called Doughnut to overcome the above problems of Rosary and LAR. By dividing nodes with the local characteristics, the Doughnut consists of Inter-Pastry and Intra-Pastry, and all nodes have the responsibility of Inter-Pastry and Intra-Pastry routing. This results in that all nodes perform the role of the existing root node. This solves the problems of the reducing of fault-tolerance, the increasing of routing hop count, and the not-distributed communication load. Also Doughnut can use cache effectively because it guarantees the even cache distribution in local(Intra-Pastry) and the cache contents in local can be used in the other local. The proposed algorithm is implemented using simulator and the experimental results show the effectiveness of the proposed method compared to the existing method.Keywords:Overlay Network, Locality, Caching, Pastry, Rosary, LAR, Doughnut

A Clustering based wireless internet proxy server needs a layer-7 load balancer with URL hashing methods to reduce the total storage space for servers. Layer-4 load balancer located in front of server cluster is to distribute client requests to the servers with the same contents at transport layer, such as TCP or UDP, without looking at the content of the request. Layer-7 load balancer located in front of server cluster is to parse client requests in application layer and distribute them to servers based on different types of request contents. Layer 7 load balancer allows servers to have different contents in an exclusive way so that it can minimize the total storage space for servers and improve overall cluster performance. However, its scalability is limited due to the high overhead of parsing requests in application layer as different from layer-4 load balancer. In order to overcome its scalability limitation, in this paper, we propose a distributed layer-7 load balancer by replacing a single layer-7 load balancer in the conventional scheme by a single layer-4 load balancer located in front of server cluster and a set of layer-7 load balancers located at server cluster. In a clustering based wireless internet proxy server, we implemented the conventional scheme by using KTCPVS(Kernel TCP Virtual Server), a linux based layer-7 load balancer. Also, we implemented the proposed scheme by using IPVS(IP Virtual Server), a linux-based layer-4 load balancer, installing KTCPVS in each server, and making them work together. We performed experiments using 16 PCs. Experimental results show scalability and high performance of the proposed scheme, as the number of servers grows, compared to the conventional scheme.Keyword:Clustering, Hashing, Layer-4 Load Balancer, Layer-7 Load Balancer, Scalability

In this paper, a 3D display processor embedding a programmable 3D graphics rendering engine is proposed. The proposed processor combines a 3D graphics rendering engine and a 3D image synthesis engine to support both true realism and interactivity for the future multimedia applications. Using high coherence between 3D graphics data and 3D display inputs, both pipelines are merged by sharing buffers such that a 3D display engine directly uses the output of a 3D graphics rendering engine. The merged architecture has synergetic coupling effects such as freely providing various rendering effects to 3D images and easily computing disparities without complex extraction processes. In the 3D image synthesis engine, we adopt view interpolation algorithm and propose real-time synthesis method, pixel-by-pixel process. The view interpolation algorithm reduces the number of images to be rendered, resulting in the reduction of external memory size to 64.8% compared to conventional synthesis process. The proposed pixel-by-pixel process synthesizes 3D images at 36 fps through bandwidth reduction of 26.7% and decreases internal memory size to 64.2% compared to typical image-by-image process. The 3D graphics rendering engine is programmable and supports the instruction sets of the latest 3D graphics standard APIs, Pixel Shader 3.0 and OpenGL|ES 2.0. The die contains about 1.7 M transistors, occupies 5 mm times 5 mm in 0.18 mum CMOS and dissipates 379 mW at 1.85 V.

A wireless router is a device which allows several wireless clients to share an internet line using NAT (Network Address Translation). In a school or a small office environment where many clients use multiple wireless routers, a client may select any one of wireless routers so that most clients can be clustered to a small set of the wireless routers. In such a case, there exists load unbalancing problem between clients and wireless routers. One of its result is that clients using the busiest router get poor service. The other is that the resource utilization of the whole wireless routers becomes very low. In order to resolve the problems, we propose a load sharing scheme to maximize network bandwidth utilization based on SSID(Service Set IDentifier) hiding. The proposed scheme keeps checking the available bandwidth of all the possible wireless routers in a time interval and select the most available one. If a new client appears, the most available router is visible to him or her whereas the others are not visible. This is handled by SSID hiding in the proposed scheme. We implemented the proposed scheme with ASUS WL 500G wireless router and performed experiments. Experimental results show the bandwidth utilization improvement compared to the existing method.

A contents correction signature hashing scheme for fast string matching algorithm in high performance network intrusion detection systems is proposed. The objective is to design a highly scalable pattern matching system suitable for wide spreading similar varieties of malicious behaviors in network. Simulation results show the proposed method maintains the hash table evenly compared with general hash table and shows high scalability as the increase of signatures and 130% of performance gain compared with SNORT.

Performance scalability and storage scalability become important in a large scale cluster of wireless internet proxy cache servers. Performance scalability means that the whole performance of the cluster increases linearly according as servers are added. Storage scalability means that the total size of cache storage in the cluster is constant, regardless of the number of cache servers used, if the whole cache data are partitioned and each partition is stored in each server, respectively. The Round-Robin based load balancing method generally used in a large scale server cluster shows the performance scalability but no storage scalability because all the requested URL data need to be stored in each server. The hashing based load balancing method shows storage scalability because all the requested URL data are partitioned and each partition is stored in each server, respectively. but, it shows no performance scalability in case of uneven pattern of client requests or Hot-Spot. In this paper, we propose a novel dynamic hashing method with performance and storage scalability. In a time interval, the proposed scheme keeps to find some of requested URLs allocated to overloaded servers and dynamically reallocate them to other less-loaded servers. We performed experiments using 16 PCs and experimental results show that the proposed method has the performance and storage scalability as different from the existing hashing method. Key Words:Duplication, Storage, Scalability, Request pattern, Hot-Spot, Dynamic hashing

Caching in a cache sorrel cluster environment has an advantage that minimizes the request and response tine of internet traffic and web user. Then, one of the methods that increases the hit ratio of cache is using the hash function with cooperative caching. It is keeping a fixed size of the total cache memory regardless of the number of cache servers. On the contrary, if there is no cooperative caching, the total size of cache memory increases proportional to the number of cache sowers since each cache server should keep all the cache data. The disadvantage of hashing method is that clients` requests stress a few servers in all the cache servers due to the characteristics of hashing md the overall performance of a cache server cluster depends on a few servers. In this paper, we propose the method that distributes uniformly client requests between cache servers using dynamic server information. We performed experiments using 16 PCs. Experimental results show the uniform distribution o