Smoke As Well As Mirrors: Reflecting Files At A Geographically Remote Place Without Loss Of Performance

This newspaper from FAST'09 introduces smoke as well as mirrors filesystem (SMFS) which mirrors files at a geographically remote datacenter alongside negligible acquit upon on performance. It turns out remote mirroring is a major occupation for banking systems which proceed off-site mirrors (employing dedicated high-speed 10-40Gbits optical links) to last disasters.

This newspaper is close disaster tolerance, non your everyday fault-tolerance. The fault model includes that the principal site may become destroyed, as well as roughly sporadic package losses upto 1% may direct house simultaneously equally well, yet withal no information should live lost. (Data is said to live lost if the customer is acknowledged for the update but the corresponding update/data no longer exists inwards the system.) The principal site existence destroyed may live a fighting over-dramatization. An equivalent means to solid pose down the fault model would live that the newspaper only rules out a post service hoc correction (replay or manual correction). Here is how manual correction would work: if might outage occurs as well as the organization drops roughly requests, as well as the mirror is inconsistent, thus when nosotros become the principal upward again, nosotros tin restore the lost requests from the principal as well as brand the mirror eventually-consistent. The newspaper rules that out, as well as insists that the organization doesn't lose whatever information ever.

Existing mirroring techniques

Here are the existing mirroring techniques inwards use:

Synchronous mirroring entirely sends acknowledgments to the customer subsequently receiving a reply from the mirror. Data cannot live lost unless both principal as well as mirror sites fail. This is the most dependable solution, but performance suffers because of wide-area oneway link latencies of upto 50ms.

Semi-synchronous mirroring sends acknowledgments to the customer subsequently information written is locally stored at the principal site as well as an update is sent to the mirror. This scheme does non lose information unless the principal site fails as well as sent packets driblet on the means to the mirror.

Asynchronous mirroring sends acknowledgments to the customer at nowadays subsequently information is written locally. This is the solution that provides the best performance, but it is besides the to the lowest degree dependable solution. Data loss tin direct house fifty-fifty if only the principal site fails.

Proposed network-sync mirroring

Clearly, semi-synchronous mirroring strikes a skillful residual betwixt reliability as well as performance. The proposed approach inwards SMFS is genuinely a pocket-sized improvement on the semi-synchronous mirroring. The basic sentiment is to ensure that in ane trial a package has been sent, the likelihood that it volition live lost is equally depression equally possible. They produce this yesteryear sending forrad fault correction (FEC) information along alongside the package as well as informing the sending application when FEC has been sent along alongside the data. (An representative of FEC is using Reed-Solomon fault correction.) They telephone yell upward this technique "network-sync mirroring".

This sentiment is uncomplicated as well as straightforward, but this function provides a really skillful execution of the idea. SMFS employs previous function of the authors, Maelstrom (NSDI'08), to furnish FEC for wide-area-network transmission. SMFS implements a filesystem that preserves the social club of of operations inwards the construction of the filesystem itself, a log-structured filesystem. The newspaper besides presents several real-world experiments to evaluate the performance of SMFS equally good equally its disaster tolerance. Here are ii graphs from the evaluation section.

Share this post