Load-balancing an asymmetrical distributed erasure-coded system

• US 8,819,261 B2
• Filed: 10/15/2009
• Issued: 08/26/2014
• Est. Priority Date: 10/15/2008
• Status: Expired due to Fees

• Current Assignee: Xenogenic Development Limited Liability Company (Intellectual Ventures LLC)
• Original Assignee: Aster RISK Management LLC (Intellectual Ventures LLC)
• Inventors: Zuckerman, Gal, Thieberger, Gil
• Primary Examiner(s): Jaroenchonwanit, Bunjob
• Application Number:
  US12/580,200
• Publication Number:
  US 20100094974A1
• Time in Patent Office: 1,776 Days
• Field of Search: None
• US Class Current: 709/231
• CPC Class Codes: G06F 16/1834 implemented based on peer-to-peer networks, e.g. gnutella p2p communication protocols H04L67/104
• CPC Class Codes: G06F 16/70 of video data
• CPC Class Codes: H03M 13/37 Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
• CPC Class Codes: H03M 13/3761 using code combining, i.e. using combining of codeword portions which may have been transmitted separately, e.g. Digital Fountain codes, Raptor codes or Luby Transform [LT] codes
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers
• CPC Class Codes: H04L 67/1008 based on parameters of servers, e.g. available memory or workload monitoring of computer activity G06F11/30
• CPC Class Codes: H04L 67/101 based on network conditions
• CPC Class Codes: H04L 67/1012 based on compliance of requirements or conditions with available server resources
• CPC Class Codes: H04L 67/1021 based on client or server locations
• CPC Class Codes: H04L 67/1023 based on a hash applied to IP addresses or costs
• CPC Class Codes: H04L 67/1091 Interfacing with client-server systems or between P2P systems
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]

Abstract

Load-balancing an asymmetrical distributed erasure-coded system including fractional-storage CDN servers, storing, at a high storage gain, erasure-coded fragments encoded with a redundancy factor greater than one from segments, and a plurality of assembling devices, each obtaining fragments from a subgroup of the servers. The subgroups are selected from the servers still capable of increasing their fragment delivery throughput. Wherein not all of the servers have the same fragment delivery bandwidth capability, and the storage gain of each segment on each server is usually not strictly proportional to the bandwidth capability of the server, and the aggregated throughput used by the servers to deliver fragments may approach the aggregated bandwidth capabilities of the servers.

Claims

1. A distributed system comprising:

a plurality of CDN servers and a larger number of assembling devices;

some of the CDN servers operative to have different fragment-delivery bandwidth capabilities than the other CDN servers;

each of the servers configured to store at least one erasure-coded fragment but less than the minimum data required to reconstruct erasure-coded segments of streaming contents; and

each of the assembling devices configured to obtain erasure-coded fragments usable to reconstruct the segments from a subgroup of the servers comprising between two and less than the plurality of servers;

wherein the system configured to approach the combined aggregated fragment delivery bandwidth capabilities of the servers by assigning serves to the subgroups based on the unutilized fragment delivery bandwidths of the servers, and essentially regardless of the number of fragments stored on each server per segment stored on the server.

2. The distributed system of claim 1, wherein the assembling devices are further configured to use a fragment pull protocol to obtain the fragments, each fragment pull request associated with a specific segment, and to select the subgroups of servers from which they obtain fragments using the fragment pull protocol.

3. The distributed system of claim 2, wherein the assembling devices select for the subgroups the servers with the highest unutilized bandwidth, and the erasure-coded fragments support source-selection diversity.

4. The distributed system of claim 2, wherein the assembling devices are further configured to detect increased latency in servers approaching their bandwidth usage limits, and to select for the subgroups other servers having lower latencies.

5. The distributed system of claim 1, wherein the assembling devices configured to reselect the subgroups from time to time while obtaining the fragments.

6. The distributed system of claim 1, wherein the assembling devices are configured to obtain the fragments using a fragment pull protocol, each fragment pull request associated with a specific segment.

7. The distributed system of claim 6, wherein the assembling devices are further configured to detect increased latency variance in servers approaching their bandwidth usage limits, and to select for the subgroups other servers having lower latency variances.

8. The distributed system of claim 6, wherein the outgoing bandwidth of at least some of the CDN servers is shared by several applications, and the CDN servers are further configured to vary their fragment delivery bandwidth capabilities accordingly.

9. The distributed system of claim 6, wherein at least some of the servers are further configured to change their fragment delivery bandwidth capabilities on-the-fly.

10. The distributed system of claim 1, wherein the assembling devices are configured to obtain the fragments using multiple streams from multiple servers, each stream contains less than a third of the required fragments, and the erasure-coded fragments support source-selection diversity.

11. The distributed system of claim 1, wherein the erasure coding is rateless-coding resulting in fragments having a redundancy factor >

4.

12. The distributed system of claim 1, wherein the servers belong to at least two different classes;

the first class comprises high bandwidth CDN servers directly connected to the Internet backbone, and the second class comprises lower bandwidth CDN servers not directly connected to the Internet backbone.

13. A distributed system comprising:

a plurality of CDN servers and a larger number of assembling devices;

some of the CDN servers operative to have different fragment-delivery bandwidth capabilities than the other CDN servers;

the CDN servers configured to store erasure-coded fragments associated with segments;

each of the servers configured to store at least one erasure coded fragment but less than the minimum data required to reconstruct the segments;

and a plurality of assembling devices, each configured to obtain at least a majority of fragments required to reconstruct the segments from a subgroup of the servers comprising between two and less than the plurality of servers;

wherein the system configured to deliver fragments at a bandwidth approaching the combined fragment delivery bandwidth capabilities of the servers, by selecting the servers of the subgroups based on a random or pseudo random selection algorithm weighted according to the unutilized bandwidth of the servers and essentially regardless of the number of fragments stored on each server per segment stored on the server.

14. The distributed system of claim 13, wherein the assembling devices are configured to obtain the fragments using a fragment pull protocol, each fragment pull request associated with a specific segment.

15. A method comprising:

distributing erasure-coded fragments associated with segments among servers not having the same fragment delivery bandwidth capabilities;

after distributing the fragments, each of the servers receives and stores less than the minimum amount of fragments needed to reconstruct the segments, and the amount of fragments received and stored by each of the servers is essentially not a function of the fragment delivery bandwidth capability of the server;

selecting, for assembling devices, subgroups of servers having enough unutilized bandwidth to provide fragments;

and approaching the combined fragment delivery bandwidth capabilities of the servers by pulling, by the assembling devices, erasure-coded fragments usable to decode segments, each fragment pull request associated with a specific segment, wherein each assembling device is retrieving the fragments from each server essentially independently of the bandwidth capability of the server.

16. The method of claim 15, wherein the erasure-coding is rateless-coding.

17. The method of claim 15, wherein the segments belong to streaming contents, and further comprising changing the servers in the subgroups from time to time while retrieving the fragments.

18. The method of claim 15, further comprising compensating for on-the-fly changes in unutilized fragment delivery bandwidth of at least some of the servers by retrieving the fragments from the servers having enough unutilized bandwidth to provide the requested fragments.

19. The method of claim 15, wherein the distributing the fragments comprises distributing the fragments to at least two different classes of servers;

the first class comprises high bandwidth CDN servers directly connected to the Internet backbone, and the second class comprises lower bandwidth CDN servers not directly connected to the Internet backbone.

20. The method of claim 15, further comprising selecting, by the assembling devices, the servers having the highest amount of unutilized fragment delivery bandwidth as their subgroups.