WO2004051943A1

WO2004051943A1 - Nas load distribution system

Info

Publication number: WO2004051943A1
Application number: PCT/JP2002/012787
Authority: WO
Inventors: Yoshitake Shinkai
Original assignee: Fujitsu Limited
Priority date: 2002-12-05
Filing date: 2002-12-05
Publication date: 2004-06-17
Also published as: JP3949688B2; JPWO2004051943A1

Abstract

A NAS load distribution system capable of automatically routing a request issued by an NAS client to a particular node, thereby distributing load between actual nodes without having a dispatcher in an NAS apparatus. In order to achieve this object, in the NAS load distribution system for distributing the NAS load of the NAS apparatus consisting of a plurality of nodes accessed via a network by the NAS client, the NAS apparatus has a plurality of virtual IP addresses.

Description

NAS load balancing system

SUMMARY OF THE INVENTION The present invention provides a NAS client file storage (NAS) device comprising a plurality of nodes (for example, computers) designed to allow the same file to be accessed from any node. The request issued by the NAS client is automatically routed to a specific node, taking into account the locality, and load distribution between real nodes is improved, and operability is improved. And technology for building NAS systems that can improve performance. Background art

Figure 1 shows a conventional NAS system in which NAS devices and NAS clients are connected via a network. The NAS system shown in FIG. 1 includes a NAS 100, a network 110, and computers 111 and 112 113 that constitute a group of NAS clients.

The NAS device 100 includes a storage device 104, a plurality of nodes 101, 102 and 103, and a dispatcher device 105. The dispatcher device 105 is known as a method for balancing the load of each of the nodes 101, 102, and 103.

In this method, one logical IP address (LIP1) is defined for the NAS device 100, and the NAS clients 111, 112, and 113 execute a file operation request to this logical IP address. As described above, in the NAS device 100, a special device for performing IP address conversion, which is a dedicated computer called the above-described dispatcher device 105, is arranged. The dispatcher device 105 assigns a logical IP address, which is the destination of the request issued by the NAS client 111, 112, or 113, to each node having a NAS device. To a different physical IP address (MAC1, MAC2 or MAC3). This causes the dispatcher device 105 to route the request to the destination node.

In the conventional; ^ :, the NAS client can access ^ ft without knowing the configuration of the nodes in the NAS device, but at the same time, it increases the device cost, and furthermore, the NAS client has an ano-head associated with the IP address conversion. There is a performance disadvantage that affects the processing power of the equipment. For example, if the frequency of requests arriving at the NAS device from the NAS client exceeds the processing capacity of the dispatcher device 105, it is not possible to increase the processing capacity in a scalable manner, regardless of the number of nodes in the NAS device. There is a problem.

FIG. 2 shows a NAS system in which a conventional NAS device and a group of NAS clients are connected via a network when the dispatcher device 105 of FIG. 1 is not used. In FIG. 2, components denoted by the same reference numerals as those in FIG. 1 indicate the same components.

In FIG. 2, the NAS device 100 includes a plurality of nodes 101, 102 and 103, and a device 104. Each node 101, 102 and 103 has a unique IP address (IP1, IP2, and IP3). When the computers 111, 112 and 113 constituting the NAS client group access the NAS device 100, the computers 111, 112 and 113, among the IP addresses (IP1, JP2 and IP3) assigned to the nodes 101, 102 or 103, The storage device 104 in the NAS device 100 is always accessed via the same one node by using any one of the above. For example, the NAS client 111 always accesses the storage device 104 via the node 101, the NAS client 112 always accesses the storage device 104 via the node 102, and the NAS client 113 always accesses the node 103. The storage device 104 is accessed via the server.

However, as described above, if each NAS client always accesses via the same node in the NAS device 100, there is a problem that 9% of the load on each node cannot be achieved. Disclosure of the invention

The present invention has been made in view of the above points, and in a NAS device composed of a plurality of nodes, a request issued from a NAS client without having a dispatcher device in the NAS device is transmitted to a specific node. The purpose of the present invention is to provide a NAS load balancing method that can achieve load balancing between real nodes by automatically routing to a network.

In order to achieve this object, the present invention has a NAS device and a plurality of nodes in a NAS device. Each node has a NAS body, a statistical data collection structure, and a destination structure, respectively, and one node also has an actual IP structure.

The NAS itself processes the file operation request issued from the NAS client and returns a response.

The statistical data collection compiles statistics of requests issued from NAS clients for each destination logical IP address. An asymmetric NAS device (a type of device that has the most suitable processing node (primary node) attribute for each file and increases processing costs when processed by other nodes after processing by the primary node) Calculate also for each destination logical IP address and for each primary node.

The physical IP configuration evaluates the data compiled by the statistical data collection system at regular intervals, and changes the NAS client's holding destination via the destination switching mechanism.

The destination switching mechanism notifies the NAS client of the real MAC address corresponding to the logical IP address of the NAS device held by the NAS client to the NAS client without using special dispatch nodeware. The actual IP address of the request to be issued is switched on the NAS client side.

According to the present invention, a request issued from a NAS client is automatically routed to a specific node using each of the above-described mechanisms, so that a dispatcher device or the like can be used without using a dispatcher device or the like. Thus, the load can be distributed among the actual nodes. BRIEF DESCRIPTION OF THE FIGURES Other objects, features and aspects of the present invention will become more apparent from the following detailed description when read in conjunction with the accompanying drawings.

FIG. 1 is a diagram showing a conventional NAS system in which a NAS device and a group of NAS clients are connected via a network.

FIG. 2 is a diagram showing a NAS system in which a conventional NAS device having no dispatcher in the NAS device and a group of NAS clients are connected via a network. FIG. 3 is a diagram illustrating a NAS system in which the NAS device and the NAS client group according to the embodiment of the present invention are connected via a network.

FIG. 4 is a diagram showing a flowchart of the operation of the NAS body of the NAS device according to the present invention.

FIG. 5 is a flowchart showing the operation of the statistical data collection mechanism of the NAS device according to the present invention.

FIG. 6 is a diagram showing a flowchart of the operation of the actual IP determining mechanism of the NAS device according to the present invention.

FIG. 7 is a diagram showing a flowchart of the operation of the destination change mechanism of the NAS device according to the present invention. BEST MODE FOR CARRYING OUT THE INVENTION

An embodiment for carrying out the present invention will be described below with reference to the drawings. FIG. 3 is a diagram showing a NAS system in which the NAS device 100 and the NAS clients 111, 112, and 113 constituting a NAS client group are connected by a network 110 according to the embodiment of the present invention. The NAS clients 111, 112, and 113 constituting the NAS client count are computers that share and use the NAS device 100.

The NAS device 100 includes a plurality of nodes 101, 102 and 103, and a storage device 104. Each of the nodes 101, 102, or 103 has a NAS body 121, 131, and 141, a statistical data structure 122, 132, and 142, and a destination change structure 123, 133, and 143, respectively. Each of the nodes 103 further has a physical I address determination mechanism 144. The outline of the operation of each component will be described below.

First, the NAS clients that make up the NAS client group will be described.

Each of the NAS clients 111, 112, and 113 is composed of a computer widely distributed in general and an existing program running on the computer. Each of the NAS clients 111, 112, and 113 is assigned a logical IP address, which is a representative IP address of the NAS device 100 as viewed from each computer, and an IP—X (X is a number determined for each NAS client).

Each NAS client 111, 112 and 113 accesses the NAS device 100: In ^, each NAS client 111, 112 and 113 speaks a request towards its respective logical IP address. The logical IP address corresponds to one of the MAC addresses or the physical addresses (MAC1, MAC2, MAC3) of each of the nodes 101, 102, and 103 of the NAS device 100. Then, the request from each of the NAS clients 111, 112, and 113 is sent to a node having a MAC address associated with each logical IP address. This association is performed in the same manner as a general mechanism for associating real IP addresses with MAC addresses.

Next, the NAS bodies 121, 131, and 141 of each node 101, 102, or 103 in the NAS device 100 will be described.

The NA> S main units 121, 131, and 141 are programs for processing a file operation request sent from the NAS client 111, 112, or 113. Each node 101, 102 and 103 of the NAS device 100 has the following structure: ^ indicates that there is no difference between the nodes 101, 102 and 103 of the NAS device 100. 111, 112 or 113 can process all files at the same cost via any of the nodes 101, 102 and 103.

On the other hand, each of the nodes 101, 102, and 103 of the NAS device 100 determines which node among the nodes 101, 102, or 103 of the NAS device 100 depends on the file to be processed. Access via More appropriate, a set of pre-defined files can be processed with minimal cost, while other files are configured to cost extra. It is assumed that this ^ is recorded in the metadata (node number) that can be processed at the lowest cost for each file. The present invention is adjusted so that it can be applied even if it has any of the above-described configurations. Next, the statistical data collection structures 122, 132, and 142 of each of the nodes 101, 102, or 103 in the NAS device 100 will be described. Each of the statistical data collection structures 122, 132, and 142 accumulates performance data notified from each of the NAS units 121, 131, and 141, and accumulates statistical data for each logical IP address. Then, at regular time intervals, the result of the accumulated statistical data is sent to the real IP decision Em structure 144 provided in the node 103.

Next, the actual IP decision structure 144 provided in the node 103 will be described. The real IP determining mechanism 144 of the node 103 further compiles the statistical data sent from all the nodes 101, 102, and 103 included in the NAS device 100 from the statistical data collection structures 122, 132, and 142. Then, at regular time intervals, a “logical IP address-real IP address conversion table” indicating the correspondence between the optimal logical IP address and the real IP address is created, for example, according to the algorithm described later. This refers to the destination change mechanism of each of the nodes 101, 102, and 103 in 00.

Next, the destination changing mechanisms 123, 133, and 143 of each of the nodes 101, 102, or 103 in the NAS device 100 will be described. The destination change mechanism 123, 133, and 143 change the destination of the logical IP address according to the optimal “logical IP address-to-real IP address conversion table” described by the real IP resolution provided in node 103. The processing of is performed.

Next, referring to FIG. 4 to FIG. 7, each of the nodes 101, 102, or 103 of the above-described NAS device 100 has the NAS main bodies 121, 131, and 141, and the gun gauge data collection it structures 122, 132, and 142. The algorithm of the operation of the physical IP address resolution structure 144 of the node 103 will be described in detail. First, the operation of the NAS bodies 121, 131, and 141 will be described. FIG. 4 shows an algorithm of the operation of the NAS bodies 121, 131 and 141. 4, the operation of the NAS bodies 121, 131, and 141 starts in step S401.

Next, in step S402, for example, the NAS 121 receives a file operation request from any of the NAS clients 111, 112, or 113 via the network 110. For example, in FIG. 3, a file operation request is received from the NAS client 112 via the network 110.

Next, in step S403, each NAS main body 121, 131, or 141 obtains a file to be processed for each reception request.

Next, in step S403, it is determined whether or not the primary node (no de-P) of the file of ¾ ^ exists. As described above, the primary node is the processing node that has the lowest processing cost and is optimal for each file in an asymmetric NAS device. ¾ If the primary node number (node_P) of the file does not exist ^, the process proceeds to step S405. If the primary node number (node-P) of the file exists: ^, the process proceeds to step S406. Next, in step S405, the own node number is substituted for the primary node number (no de-P).

Next, in step S406, the collected statistical data is sent to each of the statistical data collecting mechanisms 122, 132 and 142. The collected statistical data includes the logical IP address of the destination from the NAS client, the IP address of the NAS client that made the request, and the primary node number (or symmetric: ^ indicates the node of the node that received the request) Number).

Next, in step S407, the NAS bodies 121, 131, and 141 execute a file operation requested by the NAS client for each node. Then, finally, in step S408, the NAS transmits a response to the file operation requested by the NAS client to the requesting NAS client. After this, the NAS itself waits for a file operation request to receive a request from the next NAS client. Next, the operation of the statistical data collection structures 122, 132 and 142 will be described. FIG. 5 shows the algorithm of the operation of the gun gauge data collection β structures 122, 132 and 142.

In the algorithm of FIG. 5, in step S501, the operations of the gun gauge data collection mechanisms 122, 132, and 142 start.

Next, in step S502, each of the statistical data collection mechanisms 122, 132, and 142 compiles the statistical data sent from each of the NAS units for each logical IP address and each primary node number (no dep). accumulate. That is, the number of file operation requests is accumulated as statistical data for each logical IP address and for each primary node number (no de-P). At the same time, the number of all file operation requests is accumulated as statistical data.

Next, in step S 503, it is determined whether or not a predetermined time has elapsed before sending the statistical data collected last time to the actual IP scheme 144 provided in the node 103. If it is determined that the predetermined time has not elapsed, the process proceeds to S506 and ends. If it is determined that the predetermined time has elapsed, the process proceeds to step S504. '

Next, in step S 504, the statistical data compiled this time is transmitted to the actual IP decision structure 144 set in the node 103.

Next, in step S505, the current time at which the statistical data collected this time is transmitted to the actual IP determining mechanism 144 provided in the node 103 is set as a new total time of the statistical data.

Then, in step S506, the process ends.

Next, the operation of the actual IP determination mechanism 144 will be described. FIG. 6 shows an algorithm of the operation of the actual IP determination mechanism 144. In particular, in the present embodiment, this algorithm will be described below for the case of the asymmetric type. In this algorithm, statistical data sent from the statistical data collection structure of each node is further aggregated, and at regular time intervals, an optimal “logical IP address-to-real IP address conversion table” is created. Send it back to the destination change mechanism.

In the algorithm of FIG. 6, in step S601, the real IP Operation starts.

Next, in step S 602, the real IP determination mechanism 144 uses the statistical data collection structures 122, 132, and 142 of each of the nodes 101, 102, or 103 in the NAS device 100 to explain using FIG. Received statistical data.

Next, in step S603, the total number of destination requests is calculated for each logical IP address and for each primary node number (no de-P).

Next, in step S604, it is determined whether or not the force has passed a predetermined time since the previous aggregation. If it is determined that the predetermined time has not elapsed since the previous aggregation, the process returns to step S602, and the reception of the statistical data is continued. On the other hand, if it is determined that the predetermined time has elapsed since the previous aggregation, the process proceeds to step S605.

Next, in step S605, it is determined whether or not processing has been completed for all logical IP addresses. If it is determined that the processing has been completed; return to step S602 to continue receiving statistical data. On the other hand, if it is determined that there is an unprocessed logical IP address, the process advances to step S606 to perform processing on the logical IP address.

Step S606 includes steps S607, S608, and S609.

In step S607, the total number of requests for the logical IP address (LIP-i) is obtained, and this is set as req-tota1.

Next, in steps S 608 and S 609, among the requests to the logical IP address (LIP-i), the primary node number (ne xt -pr) receiving the largest number of requests and the maximum number of requests (maxreq-no). That is, in S608, it is determined whether or not the force satisfies the condition of (maX-req-no> TOTAL [LIPi, p-node]). If the condition (maX—req_no> TOTAL [LIPi, p-node]) is satisfied, the process proceeds to step S610.

If the condition is not satisfied, the process proceeds to step S609.

Next, in step S609, the destination of the logical IP address (LIP_i) is changed to the primary node number (next-pr). At the same time, the maximum number of requests Let (max x e qn o) be the total value of the number of requests for the r imary node ∑ TOTAL [LIP i, p—no de].

Next, in step S610, cu r-pr is set as the node (L IP-i) to which the logical IP address is currently assigned.

Next, in step S611, it is determined whether or not the two primary node numbers cu r—pr and neXt−r are not equal. If cur—pr and next—pr are equal to ^^, the process proceeds to step S605. On the other hand, when cu_pr is not equal to next—pr, the process proceeds to step S612. Next, in step S612, it is determined whether or not a force that satisfies (max-req-no> 1/2 * req-tota1) is satisfied. If this condition is not satisfied, processing proceeds to step S605. On the other hand, if this condition holds, the process proceeds to Step S613.

Next, in step S613, the logical IP (LIP-i) of the logical IP (LIP-i) is transmitted to the destination changing mechanism of each of the nodes 101, 102, or 103 from the actual configuration of the node 103. An instruction is given to change the destination from the current primary node number cur_pr to the primary node number next-pr described above. Then, the process advances to step S605 to continue processing of the next logical IP address.

As described above, this algorithm is described below for the case of the asymmetric type. On the other hand, the number of requests is calculated for each primary node, and if there is a large gap between them, the logical IP address-real IP address pair is calculated as follows: And send it to the destination change mechanism.

In this algorithm example, in order to simplify the explanation, we try to collect requests from NAS clients as much as possible at the node with the lowest processing cost. However, it is easy to improve the algorithm so that the processing load of each node is flattened in consideration of the processing cost.

Next, the operation of the destination changing mechanisms 123, 133 and 143 will be described. FIG. 7 shows an algorithm of the operation of the destination changing mechanisms 123, 133 and 143. In FIG. 7, in step S701, the destination change mechanism starts the process of changing the destination of the logical IP address.

Next, in step S702, a message is sent to the node using cu r -pr to suspend all sessions using the logical IP address (LIP_i). Stopping here means stacking requests addressed to the logical IP address (LIP-i).

Next, in step S703, in order to switch the destination of the specified logical IP address and IP-i to the MAC address corresponding to neXt_pr, (MAC address-one LIP-i pair) is used as an argument. Send an IP switching message to the network. Then, the optimum logical IP address-real IP address conversion table after the change is sent to the NAS client, and the NAS client can know that the real IP address corresponding to the logical IP address has been changed. After that, the NAS device can be accessed with this real IP address. However, the upper layer of the NAS client can access the NAS device with the old logical IP address.

This logical IP address can be switched by using a well-known method such as gratuit ou s ARP (address resolution protocol;; · how to use fe etc.) Gratuitous ARP is an ARP response when there is no ARP request. That is, if the ARP response is addressed to the broadcast hardware address ^^, all hosts on the network will receive the ARP response and refresh the ARP cache, such as refreshing the ARP cache. It is.

Next, in step S704, all sessions are closed toward LIP_i from all NAS clients. As a result, the suspended request is discarded, and the NAS client is notified that the session has been terminated. When this notification is received, the NAS client re-establishes the session for LIP-i and retransmits the file operation request, so that the request is directed from the NAS client to the new node next-pr.

Then, finally, in step S705, the destination change mechanism ends.

Claims

The scope of the claims

1. In a NAS load distribution system that distributes the NAS load of a NAS device composed of multiple nodes, which is accessed via a network from a NAS client, the NAS device is connected to multiple virtual devices. NAS load balancing system with a unique IP address.

2. The NAS load balancing system according to claim 1, further comprising means for statically assigning one of a plurality of virtual IP addresses to each fiflBNAS client.

3. The ttf! BNAS device has a plurality of logical IP addresses, and allows the NAS client to select any one of the plurality of logical IP addresses so that the NAS client can select the logical IP address. A NAS load balancing system, comprising means for performing load balancing based on localization of file access.

4. The NAS device comprising a plurality of nodes, wherein the NAS client has means for transmitting a file operation request to a tiff self-virtual IP address. The described NAS load balancing system.

.

5. The NAS device comprising a plurality of nodes, characterized in that the NAS device has means for dynamically converting a ttrlB virtual IP address into a real I address corresponding to each node. 2. The NAS load balancing system according to 1.

6. In a NAS device composed of a plurality of nodes accessed from a NAS client via a network, a means for collecting statistical information of a file to be accessed is provided for each client accessing the NAS device. NAS load balancing system characterized by having.

7. The IP address conversion means performs optimization based on the locality of file access of the NAS client using the statistical information according to claim 6, and 6. The NAS load balancing method according to claim 5, wherein the IP address is converted to a real IP address of an optimal node.

8. Self-Nas According to the access pattern of each NAS client, the 'ΙίΠΕ virtual IΡ address is converted to be directly switched to the real I の address of a highly localized node, The NAS load balancing method according to claim 7.

9. If it is necessary to change the translation of the address to a virtual address I, such as the address of the host, I will change the operation of the TC PZ I according to the change of the translation. The method according to claim 7, wherein the s client reconnects the tfria session.