Home » Dell Manuals » Servers » Dell DX6004S » Manual Viewer

Dell DX6004S DX Object Storage Application Guide - Page 19

Multi-Threading, 8. Using Persistent Connections

Add to My Manuals
Save this manual to your list of manuals

Page 19 highlights

3.7. Multi-Threading DX Storage is a multi-threaded, multi-node cluster, which means that every node in a DX Storage cluster can establish and maintain connections with many different client applications at the same time. Normally, an application opens only one SCSP connection to the cluster and sends requests and receive responses in a sequential manner. However, a single client application might choose to open more than one connection to DX Storage to achieve better response times and read/write throughput for high volume applications. Because DX Storage automatically load balances requests by causing them to be redirected to a less busy node in the cluster that is capable of servicing the request, this multi-threaded client strategy can be very effective in improving overall performance when necessary, because each client thread (or process) can be connected to different nodes within the cluster. 3.8. Using Persistent Connections The DX Storage software implements HTTP/1.1 persistent connections. That means a client application need not close the socket/connection after each request. DX Storage will hold open connections and allow the client to continue sending requests and receiving responses until either the client closes the connection explicitly, or it stops sending requests for some period of time. However, the client must close its connection and reopen a new one whenever a DX Storage response includes the header Connection: close. This is typically done when there is an error that would cause confusion as to meaning of the remaining bytes sent over the connection. A client may choose to maintain only one open connection at a time, by closing the old connection before opening the new, redirected one. Or, it may choose to maintain a pool of connections to several of the nodes in a cluster. For smaller clusters, this last approach can considerably improve response times because the client eventually has open connections to all the nodes in the cluster. When connected to very large DX Storage clusters, the client application should take care not to exceed the operating system limits on number of simultaneously open connections. Copyright © 2010 Caringo, Inc. All rights reserved 14 Version 5.0 December 2010

Section	Page
DX Object Storage Application Guide	1
Table of Contents	3
Chapter 1. Introduction to DX Storage Objects	6
1.1. Introducing Named Objects	6
1.2. Summary of Named Objects and Unnamed Objects	7
1.3. Common Named Object Terminology	7
1.4. About DX Storage Object Types	8
1.5. Bucket and Object Naming Rules	8
1.6. Simple Examples of Creating Buckets and Named Objects	8
1.7. Universal Resource Identifiers (URIs)	9
1.8. About Objects and Security	10
1.8.1. The Basics of DX Storage Security	10
1.8.2. Commonly Used Security Terminology	11
1.8.3. Basic Security Guidelines	11
1.9. Concepts for Unnamed Objects	11
1.9.1. Universally Unique Identifiers (UUIDs)	11
1.9.2. About Immutable Objects	12
1.9.3. About Anchor Streams	12
1.10. Document Typographical Conventions	13
Chapter 2. Connecting to a DX Storage Cluster	14
2.1. About the Primary Access Node (PAN)	14
2.2. Requirements for Node Accessibility	14
2.2.1. Using the DX Storage Software Development Kit (SDK)	14
2.2.2. Using Multicast-DNS (mDNS)	15
2.2.3. Using DNS Round Robin	15
2.2.4. Using a Pool of Static IP Addresses	16
2.2.5. Using a Single Static IP Address	16
Chapter 3. Application Best Practices	17
3.1. Using the Software Development Kit (SDK)	17
3.2. Dynamically Locating DX Storage Nodes	17
3.3. Optimizing Bucket and Domain Operations	17
3.4. Using HTTP Client Libraries	17
3.5. Handling Redirects	18
3.6. Using Range Headers	18
3.7. Multi-Threading	19
3.8. Using Persistent Connections	19
Chapter 4. Introduction to the Simple Content Storage Protocol (SCSP)	20
4.1. Mapping SCSP Operations to HTTP Methods	20
4.2. HTTP Overview	21
4.3. Requests and Responses	21
4.3.1. HTTP Response Codes	21
4.3.2. Redirect Responses	22
4.4. Formatting SCSP Commands for Named and Unnamed Objects	22
4.5. Undefined Responses from DX Storage	23
4.6. Normal Response Headers	24
Chapter 5. SCSP READ	27
5.1. Introduction to READ	27
5.1.1. Reading With Content Validation	27
5.1.2. Requestion Node Status and Cluster Capacity	27
5.2. Unnamed Object READ Details	28
5.3. Normal Responses to READ	28
5.3.1. Normal READ Responses for Named Objects	28
5.3.1.1. Normal READ Response for a Domain	28
5.3.1.2. Normal READ Response for a Bucket	29
5.3.1.3. Normal READ Response for a Named Object	29
5.3.2. Normal READ Response for Unnamed Objects	29
5.3.3. Other Normal READ Responses	30
5.4. Error Responses to READ	30
Chapter 6. SCSP WRITE	32
6.1. Introduction to WRITE	32
6.1.1. About the Expect: 100-continue Header and WRITE	33
6.1.2. About Replicate On Write	33
6.2. Named Object WRITE Details	33
6.3. Unnamed Object WRITE Details	33
6.4. Normal Responses to WRITE	33
6.5. Error Responses to WRITE	34
Chapter 7. SCSP DELETE	36
7.1. Introduction to DELETE	36
7.2. Unnamed Object DELETE Details	37
7.3. Named Object DELETE Details	37
7.4. Normal Responses to DELETE	37
7.5. Error Responses to DELETE	38
Chapter 8. SCSP UPDATE	39
8.1. Introduction to UPDATE	39
8.2. Unnamed Object UPDATE Details	39
8.3. Normal Responses to UPDATE	40
Chapter 9. SCSP COPY	41
9.1. Introduction to COPY	41
9.2. Unnamed Object COPY Details	41
9.3. Normal Responses to COPY	41
Chapter 10. SCSP APPEND	42
10.1. Introduction to APPEND	42
10.2. Unnamed Object APPEND Details	43
10.3. Normal Responses to APPEND	43
Chapter 11. SCSP INFO	44
11.1. Introduction to INFO	44
11.1.1. Getting Replica Counts	44
11.1.2. Getting Node Status and Cluster Capacity	44
11.2. Unnamed Object INFO Details	45
11.3. Normal Responses to INFO	45
11.3.1. Normal INFO Responses for Named and Unnamed Objects	45
11.3.2. Normal INFO Responses for Named Objects	46
11.3.3. Normal INFO Responses for Unnamed Objects	47
11.4. Error Responses to INFO	47
Chapter 12. Introduction to Object Security	48
12.1. About Security Roles	48
12.2. Security Realm Overview	49
12.2.1. Common Security Terminology	49
12.2.2. About Security Realms	49
12.2.3. About Realm Caching and Security	49
12.3. About Authorization Header Syntax	50
12.3.1. About Realm Names	51
12.3.1.1. About Realms and Buckets	51
12.3.1.2. About owner@ and @owner Syntax	51
12.3.1.3. About user@realm Syntax	51
12.4. Authorization Specification Examples	51
12.5. About Authorization Header Evaluation	52
12.6. Creating Realms	53
12.7. Updating the Realm on a Bucket	54
12.8. Administrative Override and Security	54
Chapter 13. Managing Security for Application Developers	55
13.1. What Application Developers Need to Know	55
13.2. About Bucket Authorization	55
13.3. Example: Creating Bucket Owners	56
13.4. Examples of Creating Buckets and Named Objects	56
Chapter 14. Managing Security for Domain Managers	59
14.1. What Domain Managers Need to Know	59
14.2. About Domain Manager Tasks and Responsibilities	59
14.3. Examples of Updating Domain and Administrator User Lists	59
14.3.1. Updating the Domain Realm	60
14.3.2. Updating the Domain Manager Realm	61
14.3.3. Updating a Realm Using PUT	62
Chapter 15. Using the Content-MD5 Metadata Header	64
Chapter 16. Using Lifepoint Metadata Headers	65
16.1. About Storage Policies	65
16.2. A Simple Lifecycle Example	65
16.3. Lifepoint/Lifecycle Specification	66
16.4. Constraint Types	66
16.4.1. Replication	67
16.4.2. Deletion	67
Chapter 17. Using the Allow Metadata Header	69
17.1. Overview of the Allow Header	69
17.2. Administrative Override	69
Chapter 18. Using Caching Metadata Headers	71
18.1. About Caching Headers	71
18.2. Using HTTP/1.0 Caching Headers	71
18.2.1. Last-Modified	71
18.2.2. If-Modified-Since	71
18.2.3. If-Unmodified-Since	71
18.2.4. Expires	72
18.3. Using HTTP/1.1 Caching Headers	72
18.3.1. ETag	72
18.3.2. If-Match	72
18.3.3. If-None-Match	73
18.3.4. If-Range	73
Chapter 19. Using Custom Metadata Headers	75
Chapter 20. Basic Content Integrity Assurance	76
20.1. Integrity Seals	76
20.2. Validating Reads	76
20.3. Application Initiated Hash Upgrading	77
Appendix A. Content Header Metadata	78
Appendix B. Sample Code and Client Library Behaviors with 100-Continue	80

Match case Limit results 1 per page

Version 5.0

December 2010

3.7. Multi-Threading

DX Storage is a multi-threaded, multi-node cluster, which means that every node in a DX Storage

cluster can establish and maintain connections with many different client applications at the same

time. Normally, an application opens only one SCSP connection to the cluster and sends requests

and receive responses in a sequential manner. However, a single client application might choose

to open more than one connection to DX Storage to achieve better response times and read/write

throughput for high volume applications. Because DX Storage automatically load balances requests

by causing them to be redirected to a less busy node in the cluster that is capable of servicing the

request, this multi-threaded client strategy can be very effective in improving overall performance

when necessary, because each client thread (or process) can be connected to different nodes within

the cluster.

3.8. Using Persistent Connections

The DX Storage software implements

HTTP/1.1 persistent connections

. That means a client

application need not close the socket/connection after each request. DX Storage will hold open

connections and allow the client to continue sending requests and receiving responses until either

the client closes the connection explicitly, or it stops sending requests for some period of time.

However, the client

must

close its connection and reopen a new one whenever a DX Storage

response includes the header Connection: close. This is typically done when there is an error that

would cause confusion as to meaning of the remaining bytes sent over the connection.

A client may choose to maintain only one open connection at a time, by closing the old connection

before opening the new, redirected one. Or, it may choose to maintain a pool of connections to

several of the nodes in a cluster. For smaller clusters, this last approach can considerably improve

response times because the client eventually has open connections to all the nodes in the cluster.

When connected to very large DX Storage clusters, the client application should take care not to

exceed the operating system limits on number of simultaneously open connections.