CN103336785A - Distributed storage method and distributed storage device based on network coding - Google Patents
Distributed storage method and distributed storage device based on network coding Download PDFInfo
- Publication number
- CN103336785A CN103336785A CN2013102197946A CN201310219794A CN103336785A CN 103336785 A CN103336785 A CN 103336785A CN 2013102197946 A CN2013102197946 A CN 2013102197946A CN 201310219794 A CN201310219794 A CN 201310219794A CN 103336785 A CN103336785 A CN 103336785A
- Authority
- CN
- China
- Prior art keywords
- data
- matrix
- data block
- row
- node
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a distributed storage method and a distributed storage device based on network coding, and belongs to the technical field of computer storage. The method and the device solve the problem that disk IO (Input/Output) of storage nodes is too large in the existing distributed storage method based on the network coding. The distributed storage method is suitable for a distributed storage system, and comprises the steps of data coding, data decoding and data repair. The distributed storage device comprises a data coding module, a data decoding module and a data repair module. According to the method and the device, when a data node is damaged, data not more than the size of a primary file D is downloaded from d data nodes, and the damaged data is repaired, so that the repair band width is reduced effectively; and randomly selected gamma coded data blocks are downloaded directly from the d data nodes, and the data blocks are not subjected to linear operation in the data nodes, so that the disk IO of the storage nodes can be reduced on the premise that the high usability of the data is ensured, and the disk IO efficiency of the data nodes is improved effectively.
Description
Technical field
The invention belongs to computer memory technical field, be specifically related to a kind of distributed storage method and device thereof of coding Network Based.
Background technology
Along with the degree of social informatization improves constantly, the generation of information presents explosive growth, and people are also constantly uprising the storage demand of information.At commercial field, the business datum amount of company increases doublely, comprises multi-medium data, electronic commerce data, Webpage search data etc.More and more huger data storage is also more and more higher to the requirement of memory technology.
Distributed memory system has following advantage with respect to traditional file system: extensibility, reliability, high-throughput and system cost are low, the cluster scale of the distributed memory system of current large-scale Internet firm inside is in rapid expansion, and carrying cost is also in continuous rising.Along with the increase of node number in the cluster, the probability of node failure is also constantly increasing.In order to guarantee the high availability of data, adopt the mode of complete data redundancy to carry out fault-tolerant processing in the existing distributed storage system usually.All adopt three replication policies as GFS (Google File System) and HDFS (Hadoop Distributed File System), namely each part file is all preserved 3 parts of copies at memory node.Three replication policies are realized simple, can guarantee that also data have higher availability, but replication policy can bring sizable space consuming to system, and storage efficiency is low.After data quantity stored in the file system reached the PB level, for guaranteeing the availability of data, a copy of every increase all means can increase by 1,000,000 grades of other carrying costs.
Correcting and eleting codes is the popularization of copy, in face of mass data, is guaranteeing that correcting and eleting codes can significantly reduce carrying cost under the fault-tolerant ability identical with replication policy.Be that the file D of M is divided into k data block: d with size
1, d
2... d
k, each data block size is M
1, produce n coded data block: C after the coding
1, C
2... C
n, each data block size is M
1, store this n data block into n different back end: D
1, D
2... D
n, as a data node D
iDuring damage, be from k available back end (D
I1, D
I2... D
Ik) k data block (C of lining download
I1, C
I2... C
Ik), the data block C that recovers to lose
i
Just used RS class correcting and eleting codes to realize reliable memory cheaply in the second generation GFS of Google (the being ColossUs) system, see " Google ' s globally-distributed database " Proceedings of the10th USENIX conference on Operating Systems Design and Implementation.USENIX Association, 2012.251-264.Microsoft has also used the correcting and eleting codes technology in the Azure of its exploitation platform.See " Erasure coding in windows azure storage " Proceedings of the2012USENIX conference on Annual Technical Conference.USENIX Association, 2012.2.But there is a problem in correcting and eleting codes when repairing the back end that damages: repair M
1The data block of size need be downloaded k * M altogether by network connection from k different node
1The data block of size, it is expensive repairing bandwidth like this.
At the problem of the reparation bandwidth of correcting and eleting codes, the reparation bandwidth that network code can reduce back end effectively when damaging, but can cause the disk I quantitative change of back end inside big.See " Network Coding for Distributed Storage Systems " Proc.of IEEE INFOCOM2007.
Summary of the invention
The invention provides a kind of distributed storage method of coding Network Based, its device is provided simultaneously, solve the excessive problem of disk I of the existing memory node of distributed storage method of existing coding Network Based, when back end damages, by network connection d (the individual data node of d 〉=k), download the data of no more than source document D size, repair the data of damaging, reduce to repair bandwidth effectively.And γ the coded data block (C that download is selected at random from d data node
E1, C
E2... C
E γ), data block is not carried out linear operation in back end, directly read from back end, can improve the disk I efficient of back end effectively.
The distributed storage method of a kind of coding Network Based provided by the present invention is applicable to distributed memory system, comprises data coding step, data decode step and data reparation step, and distributed memory system is by a namenode NS and P memory node { DS
1, DS
2, DS
3... DS
pConstitute, P 〉=3, the memory node that wherein is used for the storage file piecemeal is called back end, is n, 3≤n≤p; It is characterized in that:
(1) data coding step comprises following substep:
(1.1) deblocking:
Source document D is divided into big or small original data block D such as c piece
g, g=0,1...c-1 is for the residue raw data D of a raw data block size of less than
B, write down D earlier
BBig or small L
B, again it is used zero padding to supply and is the raw data block size, as original data block D
c
C=k * (d+1+i-k)-(i+1) * i/2, wherein, k is for recovering the required minimal data interstitial content of original, 2≤k<n; The number of data available node when d damages node for one of reparation, k≤d<n; I is the coding redundancy parameter, 0≤i≤k-1;
(1.2) redundancy encoding:
With c original data block D
aWith encoder matrix M
eCarry out Galois field 2
qInterior computing is encoded to r coded data block C
b, q=4,8,16,32 or 64; B=0,2 ... r-1; R=(d+1+i-k) * n;
Wherein, encoder matrix M
eIn matrix element a
B, gFor belonging to Galois field 2
qInteger, 0≤a
B, g≤ 2
q-1, encoder matrix M
eIt is the Fan Demeng matrix of the capable c row of r; Each C
bAll be c original data block (D
g)
G=0,1...c-1Linear combination, linear combination coefficient corresponds to encoder matrix M
eThe vectorial V of the row that b is capable
b, i.e. each C
bCorresponding encoder matrix M
eThe vectorial V of the row that b is capable
b
(1.3) generator data file D
Meta:
With encoder matrix M
eAnd parameter n, k, d, i, q and L
BBe kept at meta data file D
MetaIn;
(1.4) data storage:
With r coded data block C
bLeave n data node d in
fOn, f=0,1 ... n-1, α=d+1+i-k coded data block of each back end storage, and store a D
MetaCopy; Back end d
fData blocks stored is C
t, t=f * α, f * α+1 ... (f+1) α-1;
(2) data decode step comprises following substep:
(2.1) obtain file metadata information:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
(2.2) download available block:
Judge that whether data available node number is then data read failure less than k in n the data node, withdraw from; Otherwise select k data available node arbitrarily, comprise r in k the data node
kThe individual coded data block of=k * α=k * (d+1+i-k), corresponding encoder matrix M altogether
eMiddle r
kIndividual row vector:
From encoder matrix M
eThis r
kSelect c row vector in the individual capable vector, the square formation M that requires this c row vector to form
E1Reversible, download this c the vectorial corresponding c coded data block of row: C then
B1, C
B2... C
Bc
(2.3) redundant decode:
To described square formation M
E1Matrix inversion obtains its inverse matrix M
E1 -1, inverse matrix M
E1 -1Middle element is designated as b
Gj, line number g=0 wherein, 1 ... c-1, columns j=0,1 ... c-1; With inverse matrix M
E1 -1Do Galois field 2 with c the coded data block of downloading
qInterior computing obtains c original data block D
g,
D
gBe c coded data block C
B0, C
B1... C
B (c-1)Linear combination, the coefficient of linear combination is inverse matrix M
E1 -1The corresponding vectorial V of row
Di
(2.4) recover data:
With c the original data block D that obtains behind the redundant decode
gBy target order D under it
0, D
1... D
C-1Be written among the recovery file D0 last piece original data block D successively
C-1Only write its preceding L
BIndividual byte forms recovery file D0 in recovery file D0;
(3) data are repaired step, as a data node d
vDuring damage, v is 0,1 ... or n-1, the reparation of the coded data block of its storage comprises following substep:
(3.1) obtain file metadata information:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
The initial value of downloading data block number variable γ is set:
γ=(2×c×d)/((2×k-i-1)×i+2×k×(d-k+1));
(3.2) computational data piece restoration information comprises following process:
(3.2.1) putting cycle index variable N1=0, whether judge in n the data node data available node number less than d, is data repairing failure then, withdraws from; Otherwise the process of carrying out (3.2.2);
(3.2.2) from d data available node, select γ coded data block at random, with they corresponding codes matrix M
eγ the vectorial V of row
hBe combined as the capable c column matrix of γ Vs, h=1,2... γ; Put N1=N1+1;
(3.2.3) generate the reparation matrix M that one (d+1+i-k) row γ is listed as
r=[m
P, h], each element m wherein
P, hFrom Galois field 2
qInterior random value, p=1,2 ... (d+1+i-k), h=1,2 ... γ;
(3.2.4) set up the new coding matrix M that the capable c of r is listed as
e', M
e' by original capable vector sum newline vector V '
pConstitute, original row vector is the included coded data block corresponding codes matrix M of data available node
eIn row vector, by it at M
eIn original position be present in M
e' in, do Galois field 2
qInterior matrix M
rWith matrix V s multiplying, obtain newline vector V '
z:
With newline vector V '
pReplace encoder matrix M
eThe middle back end d that damages
vThe vectorial V of row of the α that stores a coded data block correspondence
z, z=v * α wherein, v * α+1 ... (v+1) * α-1;
(3.2.5) check described new coding matrix M
e' whether satisfy MDS character, be then to carry out substep (3.3), otherwise the process of carrying out (3.2.6);
(3.2.6) judging whether N1≤L, is then to turn over journey (3.2.2); Otherwise put N1=0, put γ=γ+1, turn over journey (3.2.2) then, maximum cycle L=1000~3000;
(3.3) update metadata file:
With metadata ` spare D
MetaIn encoder matrix M
eReplace with new coding matrix M
e', form the meta data file D after upgrading
Meta', copy it to each back end;
(3.4) repair data piece:
Download in (3.2.2) γ the coded data block (C of selection at random of institute
E1, C
E2... C
E γ), do Galois field 2
qInterior matrix M
rWith γ coded data block (C
E1, C
E2... C
E γ) computing, the data block C that obtains repairing
p':
C
p' be γ coded data block (C
E1, C
E2... C
E γ) linear combination, the coefficient of linear combination is for repairing matrix M
rThe corresponding vectorial V of row
r
(3.5) storage data block:
With the data block C that repairs
p' store on the new data available node.
Among the present invention, the data coding step is the basis, and it is mutually independently that data decode step and data are repaired step; When the user need carry out code storage to source document, file is carried out encoding operation, carry out the data coding step; When the user need download source document, file is carried out decode operation, carry out the data decode step; When a data node damages, the data block on this back end is repaired operation, carry out data and repair step.
In the described substep (1.1), according to the different values of coding redundancy parameter i, the effect difference of coding.When i=0, each back end data quantity stored α
RMinimum, α
R=M/k; When repairing the corrupt data node, in d the data available node, be γ from each back end data downloaded amount
R, γ
R=(M * d)/(k * (d-k+1)), wherein M is the size of source document.
When i=(k-1), each back end data quantity stored α
RFor:
α
R=(2 * M * d)/(2 * k * d-k
2+ k); When repairing the corrupt data node, in d the data available node, from each back end data downloaded amount γ
RMinimum, γ
R=(2 * M * d)/(2 * k * d-k
2+ k).
When 0<i<(k-1), each node data quantity stored α
RFor:
M/k<α
R<(2×M×d)/(2×k×d-k
2+k)。When repairing the corrupt data node, in d the data available node, from each back end data downloaded amount γ
RFor:
(2×M×d)/(2×k×d-k
2+k)<γ
R<(M×d)/(k×(d-k+1))。
Encoder matrix M in the described substep (1.2)
eBe the Fan Demeng matrix of the capable c row of r, satisfy ultimate range and can divide (MDS) character that namely the data of any k node can recover the data of original in n node;
From d data node, download γ the V that selects at random in the described substep (3.4)
iCorresponding codes data block (C
E1, C
E2... C
E γ), data block is not carried out linear operation at d data intranodal, directly reads from back end, can improve the disk I efficient of back end effectively.
The distributed storage devices of a kind of coding Network Based provided by the present invention is applicable to distributed memory system, comprises data coding module, data decode module and data reparation module.Distributed memory system is by a namenode NS and P memory node { DS
1, DS
2, DS
3... DS
pConstitute, P 〉=3, the memory node that wherein is used for the storage file piecemeal is called back end, is n, 3≤n≤p; It is characterized in that:
(1) data coding module comprises following submodule:
(1.1) deblocking submodule:
Source document D is divided into big or small original data block D such as c piece
g, g=0,1...c-1 is for the residue raw data D of a raw data block size of less than
B, write down D earlier
BBig or small L
B, again it is used zero padding to supply and is the raw data block size, as original data block D
c
C=k * (d+1+i-k)-(i+1) * i/2, wherein, k is for recovering the required minimal data interstitial content of original, 2≤k<n; The number of data available node when d damages node for one of reparation, k≤d<n; I is the coding redundancy parameter, 0≤i≤k-1;
(1.2) redundancy encoding submodule:
With c original data block D
aWith encoder matrix M
eCarry out Galois field 2
qInterior computing is encoded to r coded data block C
b, q=4,8,16,32 or 64; B=0,2 ... r-1; R=(d+1+i-k) * n;
Wherein, encoder matrix M
eIn matrix element a
B, gFor belonging to Galois field 2
qInteger, 0≤a
B, g≤ 2
q-1, encoder matrix M
eIt is the Fan Demeng matrix of the capable c row of r; Each C
bAll be c original data block (D
g)
G=0,1...c-1Linear combination, linear combination coefficient corresponds to encoder matrix M
eThe vectorial V of the row that b is capable
b, i.e. each C
bCorresponding encoder matrix M
eThe vectorial V of the row that b is capable
b
(1.3) generator data file D
MetaSubmodule:
With encoder matrix M
eAnd parameter n, k, d, i, q and L
BBe kept at meta data file D
MetaIn;
(1.4) data sub module stored:
With r coded data block C
bLeave n data node d in
fOn, f=0,1 ... n-1, α=d+1+i-k coded data block of each back end storage, and store a D
MetaCopy; Back end d
fData blocks stored is C
t, t=f * α, f * α+1 ... (f+1) α-1;
(2) data decode module comprises following submodule:
(2.1) obtain file metadata information submodule:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
(2.2) download the available block submodule:
Judge that whether data available node number is then data read failure less than k in n the data node, withdraw from; Otherwise select k data available node arbitrarily, comprise r in k the data node
kThe individual coded data block of=k * α=k * (d+1+i-k), corresponding encoder matrix M altogether
eMiddle r
kIndividual row vector:
From encoder matrix M
eThis r
kSelect c row vector in the individual capable vector, the square formation M that requires this c row vector to form
E1Reversible, download this c the vectorial corresponding c coded data block of row: C then
B1, C
B2... C
Bc
(2.3) redundant decode submodule:
To described square formation M
E1Matrix inversion obtains its inverse matrix M
E1 -1, inverse matrix M
E1 -1Middle element is designated as b
Gj, line number g=0 wherein, 1 ... c-1, columns j=0,1 ... c-1; With inverse matrix M
E1 -1Do Galois field 2 with c the coded data block of downloading
qInterior multiplying obtains c original data block D
g,
D
gBe c coded data block C
B0, C
B1... C
B (c-1)Linear combination, the coefficient of linear combination is inverse matrix M
E1 -1The corresponding vectorial V of row
Di
(2.4) recover the data submodule:
With c the original data block D that obtains behind the redundant decode
gBy target order D under it
0, D
1... D
C-1Be written among the recovery file D0 last piece original data block D successively
C-1Only write its preceding L
BIndividual byte forms recovery file D0 in recovery file D0;
(3) data are repaired module, as a data node d
vDuring damage, v is 0,1 ... or n-1, the reparation of the coded data block of its storage comprises following submodule:
(3.1) obtain file metadata information submodule:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
The initial value of downloading data block number variable γ is set:
γ=(2×c×d)/((2×k-i-1)×i+2×k×(d-k+1));
(3.2) computational data piece restoration information submodule comprises following unit:
Cycle index variable N1=0 is put in unit (3.2.1), whether judges in n the data node data available node number less than d, is data repairing failure then, withdraws from; Otherwise change unit (3.2.2);
γ coded data block selected, at random with they corresponding codes matrix M in unit (3.2.2) from d data available node
eγ the vectorial V of row
hBe combined as the capable c column matrix of γ Vs, h=1,2... γ; Put N1=N1+1:
Unit (3.2.3) generates the reparation matrix M that one (d+1+i-k) row γ is listed as
r=[m
P, h], each element m wherein
P, hFrom Galois field 2
qInterior random value, p=1,2 ... (d+1+i-k), h=1,2 ... γ;
The new coding matrix M that the capable c of r is listed as is set up in unit (3.2.4)
e', M
e' by original capable vector sum newline vector V '
pConstitute, original row vector is the included coded data block corresponding codes matrix M of data available node
eIn row vector, by it at M
eIn original position be present in M
e' in, do Galois field 2
qInterior matrix M
rWith matrix V s multiplying, obtain newline vector V '
p:
With newline vector V '
pReplace encoder matrix M
eThe middle back end d that damages
vThe vectorial V of row of the α that stores a coded data block correspondence
z, z=v * α wherein, v * α+1 ... (v+1) * α-1;
Unit (3.2.5) checks described new coding matrix M
e' whether satisfy MDS character, be rotor module (3.3) then, otherwise change unit (3.2.6);
Unit (3.2.6) judges whether N1≤L, and being then changes unit (3.2.2); Otherwise put N1=0, put γ=γ+1, change unit (3.2.2) then, maximum cycle L=1000~3000;
(3.3) update metadata file submodule:
With meta data file D
MetaIn encoder matrix M
eReplace with new coding matrix M
e', form the meta data file D after upgrading
Meta', copy it to each back end;
(3.4) repair data piece submodule:
Download in (3.2.2) γ the coded data block (C of selection at random of institute
E1, C
E2... C
E γ), do Galois field 2
qInterior matrix M
rWith γ coded data block (C
E1, C
E2... C
E γ) computing, the data block C that obtains repairing
p':
C
p' be γ coded data block (C
E1, C
E2... C
E γ) linear combination, the coefficient of linear combination is for repairing matrix M
rThe corresponding vectorial V of row
r
(3.5) storage data block submodule:
With the data block C that repairs
VpStore on the new data available node.
Compared with prior art, the present invention has the following advantages:
(1) guarantees to solve under the prerequisite of data high availability and use huge space consuming that replication policy brings and the increase of carrying cost in the existing distributed storage system.
(2) when data node damages, (the individual data node of d 〉=k) is downloaded the data of no more than source document M size, repairs the data of damaging, and reduces to repair bandwidth effectively by network connection d.
When (3) repairing the back end of a damage, from d data node, download γ the coded data block (C that selects at random
E1, C
E2... C
E γ), data block is not carried out linear operation in back end, directly read from back end, can improve the disk I efficient of back end effectively.
Description of drawings
Fig. 1 is the structural representation of the embodiment of the invention;
Fig. 2 is embodiments of the invention data coding step schematic flow sheet;
Fig. 3 is embodiments of the invention data decode steps flow chart synoptic diagram;
Fig. 4 repairs the steps flow chart synoptic diagram for the embodiments of the invention data.
Embodiment
The present invention is further described below in conjunction with drawings and Examples.
As shown in Figure 1, the embodiment of the invention comprises data coding step, data decode step and data reparation step, is applicable to by a namenode NS and 5 memory node { DS
1, DS
2, DS
3... DS
5The distributed memory system that constitutes, n=4 the memory node that wherein is used for the storage file piecemeal is called back end, is d
0, d
1, d
2, d
3
(1) data coding step as shown in Figure 2, comprises following substep:
(1.1) deblocking:
Be the original data block D that the raw data D of 64M is divided into c=k * (d+1+i-k)-(i+1) * sizes such as i/2=4 piece with size
g, g=0,1...3, k=2, d=3, coding redundancy parameter i=0;
D wherein
0, D
1, D
2, D
3Size be 16M; The size of raw data D is the integral multiple of c, so L
B=0;
(1.2) data coding:
With 4 original data block D
gWith encoder matrix M
eCarry out Galois field 2
8Interior computing is encoded to r=(d+i+i-k) * n=8 coded data block C
b, b=0,2 ... 7;
Wherein, encoder matrix M
eIn matrix element a
B, gFor belonging to Galois field 2
8Integer, 0≤a
B, g≤ 2
8-1, encoder matrix M
eIt is the Fan Demeng matrix of 8 row, 4 row; Each C
bAll be 4 original data block (D
g)
G=0,1...3Linear combination, linear combination coefficient corresponds to encoder matrix M
eThe vectorial V of the row that b is capable
b, i.e. each C
bCorresponding encoder matrix M
eThe vectorial V of the row that b is capable
b
(1.3) generator data file:
With encoder matrix M
eAnd parameter n=4, k=2, d=3, i=0 and L
B=0 is kept at meta data file D
MetaIn;
(1.4) data storage:
With 8 coded data block C
0, C
1, C
2... C
7Leave 4 data node d in
0, d
1... d
3On, each back end storage α=2 coded data block, back end d
uData blocks stored is C
U * α, C
U * α+1All store a D on each back end
MetaCopy;
(2) data decode step as shown in Figure 3, comprises following substep:
(2.1) obtain file metadata information:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
aAnd parameter n=4, k=2, d=3, i=0 and L
B=0;
(2.2) download available block:
Judging whether data available node number is less than 2 in 4 data nodes, is then data read failure, withdraws from; Otherwise select 2 data available node: d arbitrarily
2, d
3, d
2, d
3The coded data block that comprises in these 2 data nodes is: C
4, C
5, C
6, C
7, be total to corresponding encoder matrix M
eIn 4 row vectors: V
4, V
5, V
6, V
7These 4 row vectors are pressed V
4, V
5, V
6, V
7Order form square formation M
E1Coding M
eSatisfy MDS character, so square formation M
E1Reversible, download these 4 corresponding 4 coded data block: the C of row vector then
4, C
5, C
6, C
7
(2.3) redundant decode:
To described square formation M
E1Matrix inversion obtains its inverse matrix M
E1 -1, inverse matrix M
E1 -1Middle element is designated as b
Gj, line number g=0 wherein, 1 ... 3, columns j=0,1 ... 3; With inverse matrix M
E1 -1With 4 coded data block C that download
4, C
5, C
6, C
7Do Galois field 2
8Interior computing obtains 4 original data block D
g,
D
gBe 4 coded data block C
B0, C
B1... C
B (c-1)Linear combination, the coefficient of linear combination is inverse matrix M
E1 -1The corresponding vectorial V of row
Di
(2.4) read data:
Earlier with the original data block D that obtains behind the redundant decode
0, D
1, D
2, D
3By target order D under it
0, D
1, D
2, D
3Be written to successively among the recovery file D0, because L
B=0, last piece original data block D
3All write among the recovery file D0; Form recovery file D0 then;
(3) data are repaired step, as a data node d
0During damage, the reparation of the coded data block of its storage comprises following substep, as shown in Figure 4:
(3.1) obtain file metadata information:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n=4, k=2, d=3, i=0 and L
B=0;
(3.2) computational data piece restoration information comprises following process:
(3.2.1) putting cycle index variable N1=0, whether judge in 4 data nodes data available node number less than 3, is data repairing failure then, withdraws from; Otherwise the process of carrying out (3.2.2), back end d
1, d
2, and d
3All available, so the process of carrying out (3.2.2);
(3.2.2) from 3 data available node d
1, d
2, and d
3In select 3 coded data blocks at random, with they corresponding codes matrix M
e3 vectorial V of row
hBe combined as 3 row, 4 column matrix Vs, h=1,2...3; Put N1=N1+1;
(3.2.3) generate the reparation matrix M that (d+1+i-k)=2 row 3 is listed as
r=[m
P, h], each element m wherein
P, hFrom Galois field 2
8Interior random value, p=1,2, h=1,2,3;
(3.2.4) set up the new coding matrix M that 8 row 4 are listed as
e', M
e' by original capable vector sum newline vector V '
pConstitute, original row vector is the included coded data block corresponding codes matrix M of data available node
eIn row vector, by it at M
eIn original position be present in M
e' in, do Galois field 2
qInterior matrix M
rWith matrix V s multiplying, obtain newline vector V '
pFor:
With newline vector V '
pReplace encoder matrix M
eThe middle back end d that damages
vThe vectorial V of row of the α that stores a coded data block correspondence
z, z=v * α wherein, v * α+1 ... (v+1) * α-1:
(3.2.5) check described new coding matrix M
e' whether satisfy MDS character, be then to carry out substep (3.3), otherwise the process of carrying out (3.2.6).Check result is M
e' satisfy MDS character, so, carry out substep (3.3);
(3.2.6) judging whether N1≤L, is then to turn over journey (3.2.2); Otherwise put N1=0, put γ=γ+1, turn over journey (3.2.2) then, maximum cycle L=2000;
(3.3) update metadata file:
With meta data file D
MetaIn encoder matrix M
eReplace with new coding matrix M
e', form the meta data file D after upgrading
Meta', copy it to each back end;
(3.4) repair data piece:
Download in (3.2.2) 3 the coded data block (C of selection at random of institute
E1, C
E2, C
E3), do Galois field 2
8Interior matrix M
rWith 3 coded data block (C
E1, C
E2, C
E3) computing, with the back end d that damages
0The data blocks stored reparation of middle institute is C
p':
C
p' be 3 coded data block (C
E1, C
E2, C
E3) linear combination, the coefficient of linear combination is for repairing matrix M
rThe corresponding vectorial V of row
r
(3.5) storage data block:
With the data block C that repairs
VpStore a new data available node d into
5On.
Claims (2)
1. the distributed storage method of a coding Network Based is applicable to distributed memory system, comprises data coding step, data decode step and data reparation step, and distributed memory system is by a namenode NS and P memory node { DS
1, DS
2, DS
3... DS
pConstitute, P 〉=3, the memory node that wherein is used for the storage file piecemeal is called back end, is n, 3≤n≤p; It is characterized in that:
(1) data coding step comprises following substep:
(1.1) deblocking:
Source document D is divided into big or small original data block D such as c piece
g, g=0,1...c-1 is for the residue raw data D of a raw data block size of less than
B, write down D earlier
BBig or small L
B, again it is used zero padding to supply and is the raw data block size, as original data block D
c
C=k * (d+1+i-k)-(i+1) * i/2, wherein, k is for recovering the required minimal data interstitial content of original, 2≤k<n; The number of data available node when d damages node for one of reparation, k≤d<n; I is the coding redundancy parameter, 0≤i≤k-1;
(1.2) redundancy encoding:
With c original data block D
aWith encoder matrix M
eCarry out Galois field 2
qInterior computing is encoded to r coded data block C
b, q=4,8,16,32 or 64; B=0,2 ... r-1; R=(d+1+i-k) * n;
Wherein, encoder matrix M
eIn matrix element a
B, gFor belonging to Galois field 2
qInteger, 0≤a
B, g≤ 2
q-1, encoder matrix M
eIt is the Fan Demeng matrix of the capable c row of r; Each C
bAll be c original data block (D
g)
G=0,1...c-1Linear combination, linear combination coefficient corresponds to encoder matrix M
eThe vectorial V of the row that b is capable
b, i.e. each C
bCorresponding encoder matrix M
eThe vectorial V of the row that b is capable
b
(1.3) generator data file D
Meta:
With encoder matrix M
eAnd parameter n, k, d, i, q and L
BBe kept at meta data file D
MetaIn;
(1.4) data storage:
With r coded data block C
bLeave n data node d in
fOn, f=0,1 ... n-1, α=d+1+i-k coded data block of each back end storage, and store a D
MetaCopy; Back end d
fData blocks stored is C
t, t=f * α, f * α+1 ... (f+1) α-1;
(2) data decode step comprises following substep:
(2.1) obtain file metadata information:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
(2.2) download available block:
Judge that whether data available node number is then data read failure less than k in n the data node, withdraw from; Otherwise select k data available node arbitrarily, comprise r in k the data node
kThe individual coded data block of=k * α=k * (d+1+i-k), corresponding encoder matrix M altogether
eMiddle r
kIndividual row vector:
From encoder matrix M
eThis r
kSelect c row vector in the individual capable vector, the square formation M that requires this c row vector to form
E1Reversible, download this c the vectorial corresponding c coded data block of row: C then
B1, C
B2... C
Bc
(2.3) redundant decode:
To described square formation M
E1Matrix inversion obtains its inverse matrix M
E1 -1, inverse matrix M
E1 -1Middle element is designated as b
Gj, line number g=0 wherein, 1 ... c-1, columns j=0,1 ... c-1; With inverse matrix M
E1 -1Do Galois field 2 with c the coded data block of downloading
qInterior computing obtains c original data block D
g,
D
gBe c coded data block C
B0, C
B1... C
B (c-1)Linear combination, the coefficient of linear combination is inverse matrix M
E1 -1The corresponding vectorial V of row
Di
(2.4) recover data:
With c the original data block D that obtains behind the redundant decode
gBy target order D under it
0, D
1... D
C-1Be written among the recovery file D0 last piece original data block D successively
C-1Only write its preceding L
BIndividual byte forms recovery file D0 in recovery file D0;
(3) data are repaired step, as a data node d
vDuring damage, v is 0,1 ... or n-1, the reparation of the coded data block of its storage comprises following substep:
(3.1) obtain file metadata information:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
The initial value of downloading data block number variable γ is set:
γ=(2×c×d)/((2×k-i-1)×i+2×k×(d-k+1));
(3.2) computational data piece restoration information comprises following process:
(3.2.1) putting cycle index variable N1=0, whether judge in n the data node data available node number less than d, is data repairing failure then, withdraws from; Otherwise the process of carrying out (3.2.2);
(3.2.2) from d data available node, select γ coded data block at random, with they corresponding codes matrix M
eγ the vectorial V of row
hBe combined as the capable c column matrix of γ Vs, h=1,2... γ; Put N1=N1+1;
(3.2.3) generate the reparation matrix M that one (d+1+i-k) row γ is listed as
r=[m
P, h], each element m wherein
P, hFrom Galois field 2
qInterior random value, p=1,2 ... (d+1+i-k), h=1,2 ... γ;
(3.2.4) set up the new coding matrix M that the capable c of r is listed as
e', M
e' by original capable vector sum newline vector V '
pConstitute, original row vector is the included coded data block corresponding codes matrix M of data available node
eIn row vector, by it at M
eIn original position be present in M
e' in, do Galois field 2
qInterior matrix M
rWith matrix V s multiplying, obtain newline vector V '
z:
With newline vector V '
pReplace encoder matrix M
eThe middle back end d that damages
vThe vectorial V of row of the α that stores a coded data block correspondence
z, z=v * α wherein, v * α+1 ... (v+1) * α-1;
(3.2.5) check described new coding matrix M
e' whether satisfy MDS character, be then to carry out substep (3.3), otherwise the process of carrying out (3.2.6);
(3.2.6) judging whether N1≤L, is then to turn over journey (3.2.2); Otherwise put N1=0, put γ=γ+1, turn over journey (3.2.2) then, maximum cycle L=1000~3000;
(3.3) update metadata file:
With meta data file D
MetaIn encoder matrix M
eReplace with new coding matrix M
e', form the meta data file D after upgrading
Meta', copy it to each back end;
(3.4) repair data piece:
Download in (3.2.2) γ the coded data block (C of selection at random of institute
E1, C
E2... C
E γ), do Galois field 2
qInterior matrix M
rWith γ coded data block (C
E1, C
E2... C
E γ) computing, the data block C that obtains repairing
p':
C
p' be γ coded data block (C
E1, C
E2... C
E γ) linear combination, the coefficient of linear combination is for repairing matrix M
rThe corresponding vectorial V of row
r
(3.5) storage data block:
With the data block C that repairs
p' store on the new data available node.
2. the distributed storage devices of a coding Network Based is applicable to distributed memory system, comprises data coding module, data decode module and data reparation module.Distributed memory system is by a namenode NS and P memory node { DS
1, DS
2, DS
3... DS
pConstitute, P 〉=3, the memory node that wherein is used for the storage file piecemeal is called back end, is n, 3≤n≤p; It is characterized in that:
(1) data coding module comprises following submodule:
(1.1) deblocking submodule:
Source document D is divided into big or small original data block D such as c piece
g, g=0,1...c-1 is for the residue raw data D of a raw data block size of less than
B, write down D earlier
BBig or small L
B, again it is used zero padding to supply and is the raw data block size, as original data block D
c
C=k * (d+1+i-k)-(i+1) * i/2, wherein, k is for recovering the required minimal data interstitial content of original, 2≤k<n; The number of data available node when d damages node for one of reparation, k≤d<n; I is the coding redundancy parameter, 0≤i≤k-1;
(1.2) redundancy encoding submodule:
With c original data block D
aWith encoder matrix M
eCarry out Galois field 2
qInterior computing is encoded to r coded data block C
b, q=4,8,16,32 or 64; B=0,2 ... r-1; R=(d+1+i-k) * n;
Wherein, encoder matrix M
eIn matrix element a
B, gFor belonging to Galois field 2
qInteger, 0≤a
B, g≤ 2
q-1, encoder matrix M
eIt is the Fan Demeng matrix of the capable c row of r; Each C
bAll be c original data block (D
g)
G=0,1...c-1Linear combination, linear combination coefficient corresponds to encoder matrix M
eThe vectorial V of the row that b is capable
b, i.e. each C
bCorresponding encoder matrix M
eThe vectorial V of the row that b is capable
b
(1.3) generator data file D
MetaSubmodule:
With encoder matrix M
eAnd parameter n, k, d, i, q and L
BBe kept at meta data file D
MetaIn;
(1.4) data sub module stored:
With r coded data block C
bLeave n data node d in
fOn, f=0,1 ... n-1, α=d+1+i-k coded data block of each back end storage, and store a D
MetaCopy; Back end d
fData blocks stored is C
t, t=f * α, f * α+1 ... (f+1) α-1;
(2) data decode module comprises following submodule:
(2.1) obtain file metadata information submodule:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
(2.2) download the available block submodule:
Judge that whether data available node number is then data read failure less than k in n the data node, withdraw from; Otherwise select k data available node arbitrarily, comprise r in k the data node
kThe individual coded data block of=k * α=k * (d+1+i-k), corresponding encoder matrix M altogether
eMiddle r
kIndividual row vector:
From encoder matrix M
eThis r
kSelect c row vector in the individual capable vector, the square formation M that requires this c row vector to form
E1Reversible, download this c the vectorial corresponding c coded data block of row: C then
B1, C
B2... C
Bc
(2.3) redundant decode submodule:
To described square formation M
E1Matrix inversion obtains its inverse matrix M
E1 -1, inverse matrix M
E1 -1Middle element is designated as b
Gj, line number g=0 wherein, 1 ... c-1, columns j=0,1 ... c-1; With inverse matrix M
E1 -1Do Galois field 2 with c the coded data block of downloading
qInterior multiplying obtains c original data block D
g,
D
gBe c coded data block C
B0, C
B1... C
B (c-1)Linear combination, the coefficient of linear combination is inverse matrix M
E1 -1The corresponding vectorial V of row
Di
(2.4) recover the data submodule:
With c the original data block D that obtains behind the redundant decode
gBy target order d under it
0, D
1... D
C-1Be written among the recovery file D0 last piece original data block D successively
C-1Only write its preceding L
BIndividual byte forms recovery file D0 in recovery file D0;
(3) data are repaired module, as a data node d
vDuring damage, v is 0,1 ... or n-1, the reparation of the coded data block of its storage comprises following submodule:
(3.1) obtain file metadata information submodule:
Download the meta data file D of source document D
Meta, obtain encoder matrix M
eAnd parameter n, k, d, i, q and L
B
The initial value of downloading data block number variable γ is set:
γ=(2×c×d)/((2×k-i-1)×i+2×k×(d-k+1));
(3.2) computational data piece restoration information submodule comprises following unit:
Cycle index variable N1=0 is put in unit (3.2.1), whether judges in n the data node data available node number less than d, is data repairing failure then, withdraws from; Otherwise change unit (3.2.2);
γ coded data block selected, at random with they corresponding codes matrix M in unit (3.2.2) from d data available node
eγ the vectorial V of row
hBe combined as the capable c column matrix of γ Vs, h=1,2... γ; Put N1=N1+1;
Unit (3.2.3) generates the reparation matrix M that one (d+1+i-k) row γ is listed as
r=[m
P, h], each element m wherein
P, hFrom Galois field 2
qInterior random value, p=1,2 ... (d+1+i-k), h=1,2 ... γ;
The new coding matrix M that the capable c of r is listed as is set up in unit (3.2.4)
e', M
e' by original capable vector sum newline vector V '
pConstitute, original row vector is the included coded data block corresponding codes matrix M of data available node
eIn row vector, by it at M
eIn original position be present in M
e' in, do Galois field 2
qInterior matrix M
rWith matrix V s multiplying, obtain newline vector V '
z:
With newline vector V '
pReplace encoder matrix M
eThe middle back end d that damages
vThe vectorial V of row of the α that stores a coded data block correspondence
z, z=v * α wherein, v * α+1 ... (v+1) *+α-1;
Unit (3.2.5) checks described new coding matrix M
e' whether satisfy MDS character, be rotor module (3.3) then, otherwise change unit (3.2.6);
Unit (3.2.6) judges whether N1≤L, and being then changes unit (3.2.2); Otherwise put N1=0, put γ=γ+1, change unit (3.2.2) then, maximum cycle L=1000~3000;
(3.3) update metadata file submodule:
With meta data file D
MetaIn encoder matrix M
eReplace with new coding matrix M
e', form the meta data file D after upgrading
Meta', copy it to each back end;
(3.4) repair data piece submodule:
Download in (3.2.2) γ the coded data block (C of selection at random of institute
E1, C
E2... C
E γ), do Galois field 2
qInterior matrix M
rWith γ coded data block (C
E1, C
E2... C
E γ) computing, the data block C that obtains repairing
p':
C
p' be γ coded data block (C
E1, C
E2... C
E γ) linear combination, the coefficient of linear combination is for repairing matrix M
rThe corresponding vectorial V of row
r
(3.5) storage data block submodule:
With the data block C that repairs
VpStore on the new data available node.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310219794.6A CN103336785B (en) | 2013-06-04 | 2013-06-04 | A kind of distributed storage method based on network code and device thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310219794.6A CN103336785B (en) | 2013-06-04 | 2013-06-04 | A kind of distributed storage method based on network code and device thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103336785A true CN103336785A (en) | 2013-10-02 |
CN103336785B CN103336785B (en) | 2016-12-28 |
Family
ID=49244950
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310219794.6A Active CN103336785B (en) | 2013-06-04 | 2013-06-04 | A kind of distributed storage method based on network code and device thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103336785B (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103533055A (en) * | 2013-10-16 | 2014-01-22 | 南京大学镇江高新技术研究院 | Low-redundancy cloud storage system and realization method thereof based on network encoding |
CN103544270A (en) * | 2013-10-18 | 2014-01-29 | 南京大学镇江高新技术研究院 | Data-center-oriented generalized network coding fault-tolerant storage platform and working method for same |
CN103810061A (en) * | 2014-01-28 | 2014-05-21 | 河南科技大学 | High-availability cloud storage method |
CN103838649A (en) * | 2014-03-06 | 2014-06-04 | 中国科学院成都生物研究所 | Method for reducing calculation amount in binary coding storage system |
CN104407807A (en) * | 2014-10-24 | 2015-03-11 | 华中科技大学 | Storage and expansion method aiming at RS coding storage cluster |
CN105356892A (en) * | 2015-06-24 | 2016-02-24 | 深圳大学 | Network coding method and system |
CN105635252A (en) * | 2015-12-23 | 2016-06-01 | 浪潮集团有限公司 | Erasure code redundant backup strategy of Hadoop distributed file system (HDFS) |
CN105681425A (en) * | 2016-01-22 | 2016-06-15 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Multi-node repairing method and system based on distributed storage system |
CN106302573A (en) * | 2015-05-14 | 2017-01-04 | 杭州海康威视系统技术有限公司 | A kind of method, system and device using erasure codes to process data |
CN106788455A (en) * | 2016-11-29 | 2017-05-31 | 陕西尚品信息科技有限公司 | A kind of building method of the optimal partial repairable system code based on bag |
CN106911793A (en) * | 2017-03-17 | 2017-06-30 | 上海交通大学 | The distributed storage data recovery method of I/O optimizations |
CN107317844A (en) * | 2017-06-02 | 2017-11-03 | 深圳大学 | Distributed storage method and system based on the decodable minimum memory expense of sawtooth |
CN107689983A (en) * | 2017-07-05 | 2018-02-13 | 上海交通大学 | Cloud storage system and method based on low reparation bandwidth |
CN108141228A (en) * | 2015-10-09 | 2018-06-08 | 华为技术有限公司 | The coding of distributed memory system |
CN108536396A (en) * | 2018-04-08 | 2018-09-14 | 华中科技大学 | A kind of storage extended method based on network code |
CN108647108A (en) * | 2018-03-21 | 2018-10-12 | 长安大学 | A kind of minimum bandwidth based on cycle VFRC regenerates the building method of code |
CN110401703A (en) * | 2019-07-10 | 2019-11-01 | 东华大学 | Cloud storage data reconstruction method based on multistage network coding |
CN112699094A (en) * | 2021-03-23 | 2021-04-23 | 中国信息通信研究院 | File storage method, data retrieval method, corresponding device and system |
CN114615014A (en) * | 2022-01-29 | 2022-06-10 | 北京航空航天大学 | Data transmission method and system for multilayer distributed network |
CN114826720A (en) * | 2022-04-19 | 2022-07-29 | 中国工商银行股份有限公司 | Data storage method and device, computer readable storage medium and electronic equipment |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101309127A (en) * | 2008-07-10 | 2008-11-19 | 浙江大学 | Distributed MIMO transmission method based on network coding |
CN101488104A (en) * | 2009-02-26 | 2009-07-22 | 北京世纪互联宽带数据中心有限公司 | System and method for implementing high-efficiency security memory |
CN101741884A (en) * | 2008-11-25 | 2010-06-16 | 华为技术有限公司 | Distributed storage method and device |
US20110161655A1 (en) * | 2009-12-29 | 2011-06-30 | Cleversafe, Inc. | Data encryption parameter dispersal |
CN103051681A (en) * | 2012-12-06 | 2013-04-17 | 华中科技大学 | Collaborative type log system facing to distribution-type file system |
-
2013
- 2013-06-04 CN CN201310219794.6A patent/CN103336785B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101309127A (en) * | 2008-07-10 | 2008-11-19 | 浙江大学 | Distributed MIMO transmission method based on network coding |
CN101741884A (en) * | 2008-11-25 | 2010-06-16 | 华为技术有限公司 | Distributed storage method and device |
CN101488104A (en) * | 2009-02-26 | 2009-07-22 | 北京世纪互联宽带数据中心有限公司 | System and method for implementing high-efficiency security memory |
US20110161655A1 (en) * | 2009-12-29 | 2011-06-30 | Cleversafe, Inc. | Data encryption parameter dispersal |
CN103051681A (en) * | 2012-12-06 | 2013-04-17 | 华中科技大学 | Collaborative type log system facing to distribution-type file system |
Non-Patent Citations (1)
Title |
---|
邹强等: "一种基于对象的多媒体存储系统优化策略", 《计算机科学》, vol. 35, no. 12, 25 December 2008 (2008-12-25), pages 79 - 82 * |
Cited By (37)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103533055A (en) * | 2013-10-16 | 2014-01-22 | 南京大学镇江高新技术研究院 | Low-redundancy cloud storage system and realization method thereof based on network encoding |
CN103544270B (en) * | 2013-10-18 | 2016-11-23 | 南京大学镇江高新技术研究院 | The generalization network code fault tolerant storage platform at data-oriented center and method of work |
CN103544270A (en) * | 2013-10-18 | 2014-01-29 | 南京大学镇江高新技术研究院 | Data-center-oriented generalized network coding fault-tolerant storage platform and working method for same |
CN103810061A (en) * | 2014-01-28 | 2014-05-21 | 河南科技大学 | High-availability cloud storage method |
CN103810061B (en) * | 2014-01-28 | 2017-05-31 | 河南科技大学 | A kind of High Availabitity cloud storage method |
CN103838649B (en) * | 2014-03-06 | 2017-04-12 | 中国科学院成都生物研究所 | Method for reducing calculation amount in binary coding storage system |
CN103838649A (en) * | 2014-03-06 | 2014-06-04 | 中国科学院成都生物研究所 | Method for reducing calculation amount in binary coding storage system |
CN104407807A (en) * | 2014-10-24 | 2015-03-11 | 华中科技大学 | Storage and expansion method aiming at RS coding storage cluster |
CN104407807B (en) * | 2014-10-24 | 2017-06-27 | 华中科技大学 | A kind of storage expansion method for RS code storage clusters |
CN106302573A (en) * | 2015-05-14 | 2017-01-04 | 杭州海康威视系统技术有限公司 | A kind of method, system and device using erasure codes to process data |
CN105356892A (en) * | 2015-06-24 | 2016-02-24 | 深圳大学 | Network coding method and system |
CN105356892B (en) * | 2015-06-24 | 2018-11-16 | 深圳大学 | The method and system of network code |
CN108141228A (en) * | 2015-10-09 | 2018-06-08 | 华为技术有限公司 | The coding of distributed memory system |
CN105635252A (en) * | 2015-12-23 | 2016-06-01 | 浪潮集团有限公司 | Erasure code redundant backup strategy of Hadoop distributed file system (HDFS) |
CN105635252B (en) * | 2015-12-23 | 2019-06-25 | 浪潮集团有限公司 | A kind of Hadoop distributed file system HDFS correcting and eleting codes redundancy backup method |
CN105681425B (en) * | 2016-01-22 | 2019-01-22 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Multinode restorative procedure and its system based on distributed memory system |
CN105681425A (en) * | 2016-01-22 | 2016-06-15 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Multi-node repairing method and system based on distributed storage system |
CN106788455A (en) * | 2016-11-29 | 2017-05-31 | 陕西尚品信息科技有限公司 | A kind of building method of the optimal partial repairable system code based on bag |
CN106788455B (en) * | 2016-11-29 | 2019-11-22 | 陕西尚品信息科技有限公司 | A kind of building method of the optimal partial repairable system code based on packet |
CN106911793B (en) * | 2017-03-17 | 2020-06-16 | 上海交通大学 | I/O optimized distributed storage data repair method |
CN106911793A (en) * | 2017-03-17 | 2017-06-30 | 上海交通大学 | The distributed storage data recovery method of I/O optimizations |
CN107317844B (en) * | 2017-06-02 | 2020-11-03 | 深圳大学 | Distributed storage method and system based on minimum storage overhead of sawtooth decodable |
CN107317844A (en) * | 2017-06-02 | 2017-11-03 | 深圳大学 | Distributed storage method and system based on the decodable minimum memory expense of sawtooth |
CN107689983B (en) * | 2017-07-05 | 2021-02-12 | 上海交通大学 | Cloud storage system and method based on low repair bandwidth |
CN107689983A (en) * | 2017-07-05 | 2018-02-13 | 上海交通大学 | Cloud storage system and method based on low reparation bandwidth |
CN108647108B (en) * | 2018-03-21 | 2021-04-13 | 长安大学 | Construction method of minimum bandwidth regeneration code based on cyclic VFRC |
CN108647108A (en) * | 2018-03-21 | 2018-10-12 | 长安大学 | A kind of minimum bandwidth based on cycle VFRC regenerates the building method of code |
CN108536396B (en) * | 2018-04-08 | 2020-05-19 | 华中科技大学 | Storage expansion method based on network coding |
CN108536396A (en) * | 2018-04-08 | 2018-09-14 | 华中科技大学 | A kind of storage extended method based on network code |
CN110401703A (en) * | 2019-07-10 | 2019-11-01 | 东华大学 | Cloud storage data reconstruction method based on multistage network coding |
CN110401703B (en) * | 2019-07-10 | 2021-10-15 | 东华大学 | Cloud storage data recovery method based on multi-level network coding |
CN112699094A (en) * | 2021-03-23 | 2021-04-23 | 中国信息通信研究院 | File storage method, data retrieval method, corresponding device and system |
CN112699094B (en) * | 2021-03-23 | 2021-07-13 | 中国信息通信研究院 | File storage method, data retrieval method, corresponding device and system |
CN114615014A (en) * | 2022-01-29 | 2022-06-10 | 北京航空航天大学 | Data transmission method and system for multilayer distributed network |
CN114615014B (en) * | 2022-01-29 | 2023-03-17 | 北京航空航天大学 | Data transmission method and system for multilayer distributed network |
CN114826720A (en) * | 2022-04-19 | 2022-07-29 | 中国工商银行股份有限公司 | Data storage method and device, computer readable storage medium and electronic equipment |
CN114826720B (en) * | 2022-04-19 | 2024-01-30 | 中国工商银行股份有限公司 | Data storage method, data storage device, computer readable storage medium and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN103336785B (en) | 2016-12-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103336785A (en) | Distributed storage method and distributed storage device based on network coding | |
US10089176B1 (en) | Incremental updates of grid encoded data storage systems | |
US9998539B1 (en) | Non-parity in grid encoded data storage systems | |
US10108819B1 (en) | Cross-datacenter extension of grid encoded data storage systems | |
US10162704B1 (en) | Grid encoded data storage systems for efficient data repair | |
US9904589B1 (en) | Incremental media size extension for grid encoded data storage systems | |
CN104052576B (en) | Data recovery method based on error correcting codes in cloud storage | |
US20160182088A1 (en) | Method For File Updating And Version Control For Linear Erasure Coded And Network Coded Storage | |
CN106407050A (en) | Data storage method | |
CN105956128B (en) | A kind of adaptive coding storage fault-tolerance approach based on simple regeneration code | |
US20180293134A1 (en) | Techniques to recover data in a network storage system | |
CN103761195A (en) | Storage method utilizing distributed data encoding | |
CN106484559A (en) | A kind of building method of check matrix and the building method of horizontal array correcting and eleting codes | |
CN105518996A (en) | Data codec method based on binary reed-solomon code | |
CN105808170A (en) | RAID6 (Redundant Array of Independent Disks 6) encoding method capable of repairing single-disk error by minimum disk accessing | |
CN107153661A (en) | A kind of storage, read method and its device of the data based on HDFS systems | |
CN104375905A (en) | Incremental backing up method and system based on data block | |
CN109358980A (en) | A kind of pair of data update and single disk error repairs friendly RAID6 coding method | |
CN103838649A (en) | Method for reducing calculation amount in binary coding storage system | |
CN106027638A (en) | Hadoop data distribution method based on hybrid coding | |
CN110083487B (en) | Data locality-based method and system for eliminating fragments of reference data blocks | |
CN104915609A (en) | Data hiding method based on Lagrange interpolation method and multiple clouds | |
US10187084B2 (en) | Method of encoding data and data storage system | |
CN105933386B (en) | A kind of storage system construction method and device | |
US10198311B1 (en) | Cross-datacenter validation of grid encoded data storage systems |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |