JP2017091307A - Decentralized storage control method and decentralization storage system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a decentralized storage control method and a decentralized storage system capable of configuring a linear network code for "Information Flow graph" when parameters (n, k, d, α, and β) of decentralized storage at arbitrary points (α, β) on an α-β tradeoff curve are given.SOLUTION: A decentralized storage control method and a decentralized storage system according to the present invention are characterized in that a counter which counts every node fault is installed and when node faults reach a maximum count value, decentralized storage system are all refreshed and decentralized encoding is restarted from the beginning. Further, a process is arranged in which a data amount of an encoding vector is converted into an alphabet character of proper size at an information source node and a restoration node to be made small enough to ignore as compared with an encoding data amount of a content.SELECTED DRAWING: Figure 8

Description

  The present invention relates to an encoding technique that achieves an arbitrary trade-off between a data storage amount and a node regenerated data transfer amount in distributed storage.

1. Overview of Distributed Storage Technology FIG. 1 shows a (n, k) distributed storage system. FIG. 1 shows a case where n = 4 and k = 3. Let s be an information source, and s has contents to be saved and restored. The content to be saved and restored written as x ^B. B on the shoulder indicates that the content length is B bytes. That is, let x _i ε {0, 1,..., Q−1} for x ^B = {x ₁ , x ₂ ,..., X _B }. Here, the set {0, 1,..., Q−1} is referred to as “distributed storage system alphabet”. Also, taking the alphabet distributed storage system finite field F _q. Here, q = 2 ^m , and 1 byte = m bits (in short, x ^B εF _q ^B ). For example, it is conceivable that the packet length transmitted and received in the target system is 1 byte. Content is distributed and stored in n storages. Further, a person who restores data is abbreviated as “DC” below “Data Collector”. ). The DC accesses arbitrary k storages and receives data from each storage by α bytes. DC restores the original content x ^B by decoding the received data.

By the way, each node of the distributed storage is always at risk of failure. FIG. 2 shows a situation where node 1 has failed and node 5 is newly configured at the same site. In this specification, the term “site” refers to a place where a specific storage exists. At this time, the node 5 is required to have the same function as the node 1 before the failure. That is, after regeneration of the node in FIG. 2, DC will be capable of again accessing any of k storage, to restore the original content x ^B by receiving the data by α bytes from each storage. When a node fails, configuring a new node with the same function at the same site is called “regeneration”. Regeneration is performed by receiving regeneration data in units of β bytes from d nodes in which a new node (reference numeral 5 ′ in FIG. 2) has not failed.

  In FIG. 2, the two types of node 5 and node 5 ′ exist for the sake of convenience in constructing a network code (see Non-Patent Document 4) as a regenerated code later. Are the same. Node 5 'represents a node that receives data for regeneration, and node 5 represents a node that transmits data for restoration to DC. Note that the horizontal axis of FIG. 2 represents time, and in the figure, which node failed at each time point, from which node the regeneration data was transferred, from which node the DC acquired the decryption data, etc. Generally, it is called “Information flow graph”.

として与えられる（非特許文献３を参照。）。 The codes (n, k, d, α, β) for configuring the distributed data for regeneration described above are referred to as “regeneration codes”. The regenerated code is one of distributed codes. The regeneration code was proposed by Dimakis et al. (Non-Patent Document 3), and its basic properties have been clarified. The feature of the regeneration is that there is a trade-off between the data storage amount α and the regeneration data transfer amount β, which is described as the next proposition.
[Proposition 1]
It is assumed that distributed storage system parameters (n, k, d, α, β) are given. A DC can correctly restore B-byte content by accessing any k distributed storages, and a failed node can be correctly regenerated by accessing any d distributed storages. The necessary and sufficient conditions for

(See Non-Patent Document 3).

  An example of a curve (trade-off curve) represented by equation (1) is shown in FIG. 3 (Non-Patent Document 3). The upper side of the curve in FIG. 3 represents a theoretically realizable region. The problem is to construct a regeneration code that realizes a point close to an arbitrary point on the trade-off curve.

V. Jacobson, D.C. K. Smetters, J.A. D. Thornton, M.M. F. Plus, N.C. H. Briggs, and R.A. L. Brainard, “Networking named content”, in Proc. 5th ACM International Conference on Emergence Networking Experiences and Technologies, pp. 1-12, 2009. Hidenori Kuwamon, Masazumi Kurihara, "New Coding Methods for Distributed Storage Systems-Regenerative Codes and Pyramid Codes", Journal of the IEICE, vol. 98, no. 2, pp. 130-137, 2015. A. G. Dimakis, P.A. B. Godfrey, Y.M. Wu, M.M. J. et al. Wainwright, and K.W. Ramchandran, “Network coding for distributed storage systems”, IEEE Trans. Inf. Theory, vol. 56, no. 9, pp. 4539-4551, Sept. 2010. T. T. Ho, M.M. Medard, R.M. Koetter, D.C. R. Karger, M.M. Effros, J.M. Shi, and B.J. Leong, “A random linear network coding approach to multicast”, IEEE Trans. Inf. Theory, vol. 52, no. 10, pp. 4413-4430, Oct. 2006.

を満足する量であることをいう。 2. Difficulties in the prior art As shown in Fig. 2, "Information flow graph" is generally a directed aperiodic graph, so that it is appropriate to apply network codes to realize regenerated codes. Seem. Here, the network code is for realizing multicast transmission on a given network as described in Non-Patent Document 4. Attempts have been made to apply this network code to the realization of the regenerated code in the “Information flow graph”. However, when network coding is applied to the “Information flow graph”, the following difficulties are encountered.
[Difficulty 1] Each node of the distributed storage fails randomly. Therefore, the form of “Information flow graph” is not unique.
[Difficulty 2] Since the “Information flow graph” has evolved over time, the network can grow indefinitely. In other words, whenever a node failure / regeneration occurs, it grows as long as possible.
[Difficulty 3] Since the form of “Information flow graph” is not fixed in advance, matrix information for encoding must also be transmitted together with the content data. For this reason, when ^m of q = 2 ^m in the alphabet F _q of the distributed storage system is regarded as a block length, in order for the size of the content to approach the capacity limit of the regenerated code, an encoding matrix (or encoding vector) ) Information must be o (m) in size. Here, the amount f (m) depending on m is the size of o (m).

It is an amount that satisfies

  Due to the difficulties described above, at present, a specific code is configured only for a specific point where α and β are minimum, instead of configuring a regenerated code at an arbitrary point on the trade-off curve. (For example, refer nonpatent literature 2.). The minimum code for α is referred to as “Minimum Storage Regeneration (MSR) code”, and the minimum code for β is referred to as “Minimum Bandwidth Regeneration (MBR) code”.

  The specific configuration of codes other than the MSR / MBR points is not only distributed storage but also the contents by extending the cache function attached to each node on the network such as ICN and CCN (see Non-Patent Document 1, for example).・ It is considered that the data transfer method used as storage will become important in the future.

3. As described in the previous section, given a distributed storage parameter (n, k, d, α, β) that includes any point (α, β) on the α-β trade-off curve, It is ideal to configure the regenerated code to be realized. However, the natural approach of constructing a linear network code for “Information flow graph” causes the difficulties described in Section 2.

  Therefore, the present invention is directed to “Information flow graph” when distributed storage parameters (n, k, d, α, β) including arbitrary points (α, β) on the α-β trade-off curve are given. It is an object of the present invention to provide a distributed storage control method and a distributed storage system that can be configured with a linear network code.

  In order to achieve the above object, the distributed storage control method and distributed storage system according to the present invention are provided with a counter to be added for each node failure, and if the node failure reaches the maximum counter value, the distributed storage system is All were refreshed and distributed encoding was restarted from the beginning. Also, in order to make the data amount of the encoded vector small enough to be ignored compared to the encoded data amount of the content, the information storage node converts the distributed storage system alphabet into an appropriately sized encoded alphabet, and restore node The processing to reversely convert the encoded alphabet to the distributed storage alphabet is arranged.

Specifically, the distributed storage control method according to the present invention is:
An information source node holding content data;
A plurality of storage nodes holding distributed data in which the content data is distributed;
A restoration node that receives the distributed data output from any of the storage nodes and restores the content data;
A distributed storage control method for controlling a distributed storage system having a distributed storage system configured by:
A conversion procedure for converting a distributed storage alphabet that encodes the content data at the information source node into an encoding alphabet that is smaller than the distributed storage alphabet;
An inverse conversion procedure for returning the encoded alphabet to the distributed storage alphabet at the restoration node;
Regeneration procedure for receiving a data from at least one other storage node at the time of failure of the storage node, forming a regeneration storage node to replace the failed storage node, and updating the distributed storage system;
Counting the number of failures of the storage node, and when the number reaches the upper limit, resetting the configured distributed storage system and configuring the distributed storage system again,
It is characterized by performing.

The distributed storage system according to the present invention is
An information source node holding content data;
A plurality of storage nodes holding distributed data in which the content data is distributed;
A restoration node that receives the distributed data output from any of the storage nodes and restores the content data;
A distributed storage system having a distributed storage system comprising:
The information source node has a conversion function of converting a distributed storage alphabet encoding the content data into an encoding alphabet smaller than the distributed storage alphabet;
The restoration node has an inverse conversion function for returning the encoded alphabet to the distributed storage alphabet;
A regeneration function for receiving data from at least one other storage node at the time of failure of the storage node, forming a regeneration storage node to replace the failed storage node, and updating the distributed storage system;
Counting the number of failures of the storage node, and when the number reaches the upper limit, resetting the configured distributed storage system and configuring the distributed storage system again,
It is characterized by providing a control part provided with.

  The present invention makes it possible to make the “Information flow graph” finite in terms of time and space by introducing a maximum counter value, and to perform network coding of regenerated codes using a random network coding algorithm. realizable. Since the present invention uses a linear network code, the reproduction code can be encoded very easily by using a scalar product of vectors and decoding using a Gaussian sweeping method.

  Therefore, the present invention is based on “Information flow graph” when given distributed storage parameters (n, k, d, α, β) including arbitrary points (α, β) on the α-β trade-off curve. A distributed storage control method and a distributed storage system that can be configured with a linear network code can be provided.

  As a preferred embodiment, in the distributed storage control method according to the present invention, at least one input data that is network encoded is input to the storage node, a random network encoding algorithm is applied as a network code configuration algorithm, and the input data The output data to be output to the subsequent node is calculated by multiplying the distributed data included in the data and the encoding matrix composed of the encoding alphabet converted by the conversion procedure.

  In addition, the storage node of the distributed storage system according to the present invention receives at least one input data that is network encoded, applies a random network encoding algorithm as a network code configuration algorithm, and is included in the input data The output data to be output to the subsequent node is calculated by multiplying the distributed data and the encoding matrix composed of the encoding alphabet converted by the information source node.

  The present invention provides a linear network for "Information flow graph" given distributed storage parameters (n, k, d, α, β) including arbitrary points (α, β) on the α-β trade-off curve. A distributed storage control method and a distributed storage system that can be configured with codes can be provided.

It is a figure explaining the example of Information flow graph of a distributed storage system. It is a figure explaining the example of Information flow graph of a distributed storage system. It is a figure explaining the example of the alpha-beta trade-off curve described in the nonpatent literature 3. FIG. It is a figure explaining the storage node of the distributed storage system which concerns on this invention. It is a figure explaining the restoration | recovery node of the distributed storage system which concerns on this invention. It is a figure explaining operation | movement (assignment of an encoding vector) of the information source node of the distributed storage system which concerns on this invention. It is a figure explaining operation | movement (assignment of an encoding vector) of the storage node of the distributed storage system which concerns on this invention. It is a figure explaining the distributed storage system which concerns on this invention.

4). 4. Detailed Description of Development Technology 4.1 System Overview Embodiments of the present invention will be described with reference to the accompanying drawings. The embodiments described below are examples of the present invention, and the present invention is not limited to the following embodiments. In the present specification and drawings, the same reference numerals denote the same components.

FIG. 8 is a diagram illustrating the distributed storage system 301 of this embodiment. The distributed storage system 301
An information source node 5 holding content data;
A plurality of storage nodes 10 holding distributed data in which the content data is distributed;
A restoration node 20 that receives the distributed data output from any storage node 10 and restores the content data;
A distributed storage system having a distributed storage system, wherein the information source node 15 converts the distributed storage system alphabet encoding the content data into an encoded alphabet smaller than the distributed storage system alphabet A processing unit 16;
The restoration node 20 includes an alphabet conversion processing unit 24 that returns the encoded alphabet to the distributed storage alphabet.
A regeneration function that receives data from at least one other storage node 10 when the storage node 10 fails, forms a regeneration storage node that replaces the failed storage node 10, and updates the distributed storage system;
Counting the number of failures of the storage node, and when the number reaches the upper limit, resetting the configured distributed storage system and configuring the distributed storage system again,
It is characterized by providing a control part provided with.

The storage node 10
At least one input data subjected to network encoding is input, a random network encoding algorithm is applied as a network code configuration algorithm, and the distributed data included in the input data and the encoded alphabet converted by the information source node 15 are used. Output data to be output to the subsequent node is calculated by multiplying the configured encoding matrix. The calculation is a process of Equation (13) described later.

  The distributed storage system 301 constructs a regeneration code that achieves an arbitrary point on the α-β trade-off curve by overcoming the difficulty described in Section 2 as follows.

For Difficulty 2:
The distributed storage system 301 has a counter to be added for each node failure. When the node failure reaches the maximum counter value T, all the distributed storage systems are refreshed to perform distributed encoding (in this embodiment, network code). )) From the beginning. By setting the maximum counter value, the “Information flow graph” does not extend infinitely (in the time direction).

Regarding Difficulty 1:
By defining the maximum counter value T, the number of topologies that the “Information flow graph” can take before the failure occurs T times is limited. Therefore, if a random network encoding algorithm (see Non-Patent Document 4) is applied to the network code configuration algorithm, encoding appropriate for each link that is a communication path between nodes, that is, each side of the “Information Flow Graph”. It can be shown that the probability that a vector is allocated is as close to 1 asymptotically with respect to the byte length m of the distributed storage alphabet or the encoded alphabet. Here, “appropriate encoding vector allocation” refers to an allocation of an encoding vector so that the restoration node 20 can correctly decode any content data. “Asymptotically approaching 1 as much as possible with respect to the byte length m” means that the configuration of the encoding vector is made random as shown in equations (10) and (11), as will be described later. This means that the probability that an appropriate encoded vector can be obtained can be increased if is made sufficiently large.

Regarding Difficulty 3:
Distributed storage system 301, converted from an alphabet Fq of distributed storage systems are given in advance a small coding alphabet ^{Fq ~ = {0,1,2, ···,} q ~ -1} ( provided that ^q ~ <q) to configuring the network code in the encoding alphabet over Fq ^~ by. For this reason, the distributed storage system 301 can ignore the data amount of the coding vector for the network code in the asymptotic limit where the byte length m is sufficiently long compared to the data amount β for regeneration.

4.2 Specific Code Configuration The operation of the distributed storage system 301 is as follows:
(0) transformation into coded alphabets Fq ^~ from the distributed storage system alphabet Fq,
(1) Configuration of encoding vector,
(2) Configuration of encoding and transmission data in each link,
(3) Decoding in DC (restoration node), and inverse conversion from encoding alphabet Fq ^to distributed storage system alphabet Fq,
It consists of 4 steps.
FIG. 4 is a diagram for explaining the configuration and operation (steps (1) and (2)) of the node ν (storage node 10). FIG. 5 is a diagram for explaining the configuration and operation of the restoration node 20 (step (3)).

すなわちｘ∈Ｆ_ｑを２進表現すれば（ｂ_１、・・・、ｂ_ｍ）とｍビットで表すことができる。写像ψは、この２進表現をｍ^δビット毎に区切ることで

として表現したものである。また、ｘ_１、ｘ_２∈Ｆ_ｑに対して

として定義する。ここで“＊”は文字列の連接を表すものとする。 Step (0) [Conversion from distributed storage system alphabet to coded alphabet]
A one-to-one mapping ψ from the distributed storage system alphabet F _q (where q = 2 ^m ) to the encoded alphabet F _q ^˜ (where q ^˜ = 2 ^ε , ε = m ^δ ) is defined as follows.

That is, if x∈F _q is expressed in binary, it can be expressed as (b ₁ ,..., B _m ) with m bits. The mapping ψ is obtained by dividing this binary expression into m ^δ bits.

It is expressed as For x ₁ , x ₂ ∈F _q

Define as Here, “*” represents concatenation of character strings.

Step (1) [Configuration of Encoding Vector]
Assume that a positive constant 0 <δ <1 is given in advance. To initially determine the coding alphabet, conversion of the source node 15 (source node s) from alphabet _{F q F} ^q to ^{~ (where, q ~ = 2 ε <q} = 2 m) row (See Equations (3) and (5)). Random network coding algorithm executing unit 12 utilizes the coding vector on the encoding alphabet F _q ^~ input to the input unit 11, constituting the encoded vector on exit link e 'therefrom. Since the target “Information flow graph” is a directed aperiodic graph, the information source node s can be ordered first. In accordance with this order, encoding vectors are recursively assigned to links emanating from each node.

が割り当てられる。ここで、［１：Ｂ］＝｛１、２、・・・、Ｂ｝を表すものとする。単位容量あたり１つの符号化ベクトルが割り当てられることに注意しておく。 FIG. 6 is a diagram for explaining the operation of the information source node s. It is assumed that a link of capacity B is incident on the information source node s as shown in FIG. At this time, the encoding vector is in the incident link to s.

Is assigned. Here, it is assumed that [1: B] = {1, 2,..., B}. Note that one encoding vector is allocated per unit capacity.

が割り当てられているものとする。ここで、

である。ただしＴは転置を表す。このデータは符号語と一緒に図４の入力部１１に蓄積される。このときノードνの出射リンクｅ’に対しては

のＲ_ｅ’個の符号化ベクトルが割り当てられることになる。各々の符号化ベクトルは、

として与えられる。ただし、係数

は符号化アルファベットＦ_ｑ ^〜より一様ランダムに取り出されたものとする。ここで、Ｉｎ（ν）はノードνへ入射するリンクの集合を表す。また「Ｉｎｆｏｒｍａｔｉｏｎ ｆｌｏｗ ｇｒａｐｈ」においてリンク容量Ｒ_ｅはαもしくはβの値をとる。 FIG. 7 is a diagram for explaining the operation of the node ν. The encoding vector is used for the incident link e of the capacity R _{e at} the node ν.

Is assigned. here,

It is. However, T represents transposition. This data is stored in the input unit 11 of FIG. 4 together with the code word. At this time, for the outgoing link e ′ of the node ν

R _{e ′} encoding vectors are assigned. Each encoding vector is

As given. However, the coefficient

And ones that were retrieved coded alphabet F _q uniformly random than ^~. Here, In (ν) represents a set of links incident on the node ν. The link capacity _{R e} in the "Information flow graph" takes the value of α or β.

として与えられる。ここで、Ｏｕｔ（ν）はノードνから出射するリンクの集合を表す。 Next, in order to execute encoding, an encoding matrix is constructed from the encoding vector given by equation (10). Hereinafter, let m ^1−δ × m ^1−δ unit matrix be I. The encoding matrix on the link e∈Out (ν) emanating from the node ν is

As given. Here, Out (ν) represents a set of links emitted from the node ν.

によって実行する。このときリンクｅが出力する出力データを

とする。（１４）式には未知の量ｘ^Ｂが見かけ上含まれているが、ノードνが受け取るデータ（入力部１１が受信する入力データ）を

とすると、

によって与えられる。 Step (2) [Configuration of encoding and transmission data in each link]
The operation of the output unit 13 in FIG. 4 will be described. The encoding in link e will be described. For e∈Out (ν), encode

Run by. At this time, the output data output by link e

And (14) includes the apparent unknown amount x ^B is the expression, the node ν receives data (input data input unit 11 receives)

Then,

Given by.

ならば、適切な符号化ベクトルが各リンクに割り当てられる確率が十分大きくなることが示せる。ただし、Ｔは予め設定された最大カウンタ値である。また符号化ベクトルに関するデータを送信する際のデータ量は（１４）式よりＢＲ_ｅｍ^δビットであるので、ｍが十分大きければ符号語のデータ量Ｒ_ｅｍビットに比較して無視できるほど小さくなる。 The data sent from the node ν to the DC is sent from the node ν ′ (the physical node is the same as the node ν but is logically distinguished as the node that receives the data for regeneration). Data is transferred as it is. Where q ^~ = 2 ^mδ

Then, it can be shown that the probability that an appropriate encoding vector is assigned to each link is sufficiently large. T is a preset maximum counter value. Further, since the data amount when transmitting the data related to the coding vector is BR _e m ^δ bits from the equation (14), if m is sufficiently large, the data amount of the code word is small enough to be ignored in comparison with R _e m bits. Become.

と書く。これらのデータは入力部２１に蓄積される。ただし、Ａ^〜 _ｉａは対応する符号化ベクトルを横に並べたものをまとめて行列表現としたものである。このとき復号化は次の一次方程式

を解くことにより実行される。一次方程式の求解は掃き出し法実行部２２でなされる。ここで、Ａ_ｉａはＡ^〜 _ｉａの対応する符号化ベクトルの各要素にｍ^１−δ×ｍ^１−δ単位行列を掛けたものである。次に変換処理部２４で符号化アルファベットから分散ストレージ系アルファベットへの逆変換処理がなされる。処理内容は，ステップ（０）で説明した変換処理の逆処理である。要するに

は１対１写像であるので一次方程式（１９）式の解ψ（ｘ^Ｂ）から求めるコンテンツデータｘ^Ｂを復元できることになる。 Step (3) [Decoding in DC]
The operation of the restoration node 20 in FIG. 5 will be described.
When DC (restoration node 20) accesses k nodes ν _i1 ,..., Ν _ik , data sent from node ν _ia

Write. These data are accumulated in the input unit 21. Here, A ^to _ia are matrix representations of the corresponding encoded vectors arranged side by side. At this time, decoding is performed by the following linear equation

It is executed by solving The solution of the linear equation is performed by the sweep-out method execution unit 22. Here, A _ia is obtained by multiplying each element of the corresponding encoding vector of A ^to _{ia by} an m ^1−δ × m ^1−δ unit matrix. Next, the conversion processing unit 24 performs reverse conversion processing from the encoded alphabet to the distributed storage alphabet. The processing content is the reverse processing of the conversion processing described in step (0). in short

Since this is a one-to-one mapping, the content data x ^B obtained from the solution ψ (x ^B ) of the linear equation (19) can be restored.

5. Effects produced by the invention ICN / CCN of Non-Patent Document 1 is one of the leading methods of next-generation content distribution. This aims to reduce the network load and improve the effective throughput for the end user by redefining the cache associated with the node on the network as a local storage. Although the specific architecture of ICN / CCN is currently at the research stage, it is clear that some kind of distributed storage will be realized on the network. The amount of data stored in the distributed storage and the amount of data transferred for regeneration have a trade-off relationship expressed by equation (1). Therefore, if an optimal code cannot be constructed unless it is an MSR / MBR point, the amount of data that can be stored at each node on the network will be significantly limited, and the maximum of ICN / CCN will be imposed. It seems that it will be difficult to achieve the maximum performance. Since the present invention enables the configuration of a regenerated code at an arbitrary point on the α-β tradeoff curve, the cache (or local storage) function of ICN / CCN can be maximized. .

6). As described in section 4.1, the present invention has made it possible to make the “Information flow graph” finite in terms of time and space by introducing a maximum counter value. It is a point. As a result, even if the encoding vector coefficient is randomly assigned when using the random network encoding algorithm, the probability that the encoding vector coefficient is appropriately assigned will be close to 1 as long as the byte length of the alphabet is sufficiently large. I was able to show that. Here, the appropriate encoding vector assignment refers to an assignment of an encoding vector that allows the DC to correctly decode any content data. In addition, by using conversion to an appropriate encoding alphabet, the encoding alphabet can be made smaller than that of the distributed storage system, so that it is necessary to transfer the coefficient of the encoding vector when transferring the regenerated data. The amount of data could be ignored compared to the content data for regeneration.

10: storage node 11: input unit 12: random network encoding algorithm execution unit 13: output unit 15: information source node 16: alphabet conversion processing unit 20: restoration node 21: input unit 22: sweep-out method execution unit 24: alphabet conversion Processing unit 30: Control unit 31: Regeneration function 32: Reset function 301: Distributed storage system

Claims (4)

An information source node holding content data;
A plurality of storage nodes holding distributed data in which the content data is distributed;
A restoration node that receives the distributed data output from any of the storage nodes and restores the content data;
A distributed storage control method for controlling a distributed storage system having a distributed storage system configured by:
A conversion procedure for converting a distributed storage alphabet that encodes the content data at the information source node into an encoding alphabet that is smaller than the distributed storage alphabet;
An inverse conversion procedure for returning the encoded alphabet to the distributed storage alphabet at the restoration node;
Regeneration procedure for receiving a data from at least one other storage node at the time of failure of the storage node, forming a regeneration storage node to replace the failed storage node, and updating the distributed storage system;
Counting the number of failures of the storage node, and when the number reaches the upper limit, resetting the configured distributed storage system and configuring the distributed storage system again,
The distributed storage control method characterized by performing. In the storage node,
At least one input data that is network encoded is input, a random network encoding algorithm is applied as a network code configuration algorithm, and the distributed data included in the input data and the encoded alphabet converted by the conversion procedure The distributed storage control method according to claim 1, wherein output data to be output to a subsequent node is calculated by multiplying the configured encoding matrix. An information source node holding content data;
A plurality of storage nodes holding distributed data in which the content data is distributed;
A restoration node that receives the distributed data output from any of the storage nodes and restores the content data;
A distributed storage system having a distributed storage system, wherein the information source node converts a distributed storage alphabet encoding the content data into an encoded alphabet smaller than the distributed storage alphabet Have
The restoration node has an inverse conversion function for returning the encoded alphabet to the distributed storage alphabet;
A regeneration function for receiving data from at least one other storage node at the time of failure of the storage node, forming a regeneration storage node to replace the failed storage node, and updating the distributed storage system;
Counting the number of failures of the storage node, and when the number reaches the upper limit, resetting the configured distributed storage system and configuring the distributed storage system again,
A distributed storage system comprising a control unit comprising: The storage node is
At least one input data that is network encoded is input, a random network encoding algorithm is applied as a network code configuration algorithm, and the distributed data included in the input data and the encoded alphabet converted by the information source node 4. The distributed storage system according to claim 3, wherein output data to be output to a subsequent node is calculated by multiplying the configured encoding matrix.

Images