Geo-spreading is the distribution of data across multiple physical storage systems located in multiple locations.
Geo-Spreading with Object Storage: Erasure Coding + Multiple Sites
Unlike traditional data protection schemes that are based upon replication, object storage employs erasure coding–a data protection scheme that breaks data into shards (fragments or symbols) that are encoded with parity (redundant data), and then stored across multiple storage media. With these advanced algorithms, only a subset of the shards is needed to rehydrate the data and make it available. Data protection and hence availability are further enhanced by distribution of the shards across multiple locations. This scheme has several advantages:
- In case of a component or data center failure, data remains accessible without significantly degraded performance (as typically occurs during a RAID rebuild) since there is no such rebuild involved.
- Failed storage components can be replaced when convenient, as they do not jeopardize data availability up to the limits specified in the erasure coding.
- The storage capacity required to host the shards is a fraction of 3-way or even 2-way RAID replication approaches.
Best practice for erasure coding states that no single drive should hold more than one shard of an object, and a single node (or data center in a geo-spread configuration) never holds more shards than the object can survive losing. This ensures data accessibility even if multiple components should fail. For example, object-based storage solutions such as the Active Archive System and ActiveScale™ family can employ an eighteen symbol object sharding approach that can survive the loss of up to five shards (18/5) in a single data center deployment. Thus, access to up to five drives or nodes containing up to five required drives could be blocked without loss of data access. However, what about protection against a data center outage?
Geo-Spreading Delivers Data Protection Across Availability Zones
For most enterprises and service providers, a single location is not sufficient to deliver a high availability data protection strategy since the entire data center might go offline or its access be blocked by a natural disaster or other event. In response, organizations commonly deploy data across multiple locations or availability zones. These zones are geographically spread so they rely on discrete power sources, and will not be collectively affected by the same act of nature. If this is not feasible, at a minimum, companies will deploy across three different buildings on the same campus.
[Tweet “How to safely spread data across various geographies w/ geo-spreading | #cloud”]
For a three-data-center geo-spread deployment, an erasure coding that permits rehydration of data even if one-third (or more) of the shards become unavailable is required. This ensures that multiple component failures can be tolerated but also that a connectivity loss to an entire data center would not cause data to become inaccessible. Traditionally, the geographic spreading of data increases the complexity of data management as the volume of data is triple that of a single site and each location has to be managed separately. However, a 3-geo configuration with a single namespace means that all of the storage can be centrally managed.
The Active Archive System and ActiveScale family of solutions have been architected to support 3-geo configurations that deliver cost-effective data protection @Scale. These solutions feature:
- A choice of erasure coding (18/8, 18/7, 18/6) that allows each data center to hold only six object shards such that only those six are lost if the data center is inaccessible. In the case of 18/7 or 18/8 encoding, objects will still be available even if there are one or two additional failures respectively in the two remaining data centers.
- BitSpread optimizes distribution of data across multiple physical storage systems in multiple locations to provide for uninterrupted data availability in case of loss of an entire data center.
- BitDynamics™ protects against bit rot to ensure that the desired data will be successfully retrieved by continuously validating the object shards.
- Single namespace support for 3-geo configurations within the Active Archive System and ActiveScale family simplifies storage management by eliminating the need to configure disk or replication groups.
Why You Should Care
Protecting access to your data is a paramount concern for any organization. Things can, and do, go wrong. It’s a fact of life, sooner or later any piece of hardware in a storage system will fail. The good news with the Active Archive System and ActiveScale family is that if a drive or node fails, the system will automatically create replacement shards (symbols) for those that were lost. This is done by BitDynamics as part of its proactive scans to ensure that no at-rest corruption has occurred.
With a 3-geo configuration, the Active Archive System and ActiveScale family can deliver fifteen nines’ durability. What does that mean for your data? It means the probability is 1 in 10 trillion that you will suffer a data loss. If you are serious about data protection, this probability is hard to beat.
Managing data at scale can be a significant challenge, especially with RAID-based architectures. A three-site geographically distributed Active Archive System or ActiveScale solution delivers petabytes of object-based storage, with tens of gigabits per second of throughput, fifteen nines of durability, high availability, extreme management simplicity, and cost efficiency.
To learn more about the GeoSpreading of Objects, I suggest you read the latest white paper from analyst Alastair Cooke of The Virtualization Practice. Or check out the WD Quick FAQ.
Read the White Paper: GeoSpreading Objects with the Active Archive System
Watch the WD Quick FAQ: Triple Mirroring vs. 3-Geo OBS Configuration