Flexible, Wide-Area Storage For Distributed Systems Amongst Wheelfs

One of my students, Serafettin Tasci, wrote a practiced review of this paper, therefore I volition salve fourth dimension yesteryear using his review below, instead of writing a review myself.

In this paper the authors advise a storage arrangement for wide-area distributed systems called WheelFS. The primary contribution of WheelFS is its might of adaptation to unlike types of applications amongst unlike consistency, replica placement or failure treatment requirements. This might is obtained via semantic cues that tin endure easily expressed inwards path names. For illustration to forcefulness the primary site of a folder toilet to endure X, nosotros tin specify the cue “home/users/.Site=X/john”. This representation enables preserving of POSIX semantics as well as tyke alter inwards application software to locomote the cues.

In WheelFS, at that spot are iv groups of semantic cues. Placement cues are used to accommodate the location of primaries as well as replicas of a file or folder. Durability cues specify the disclose as well as usage of replicas. Consistency cues maintain a tradeoff betwixt consistency as well as availability via timeout limits as well as eventual consistency. And finally, large read cues are useful inwards reading large files faster via entire file prefetching as well as usage of vecino customer caches.

WheelFS consists of clients as well as servers. Clients run applications that locomote WheelFS as well as uses FUSE to acquaint the distributed file arrangement to these applications. In addition, all clients pick out local caches. Servers come about file as well as directory objects inwards storage devices. They grouping objects into structures called slices.

A 3rd gene of WheelFS is the configuration service which keeps while tables that comprise object-server assignments. Each entry inwards the while tabular array contains a replication policy as well as replicas for a slice. Configuration service is replicated on a modest ready of servers as well as uses Paxos for original election. It also provides a locking interface to servers yesteryear which the usage of slices as well as while tabular array yesteryear servers is coordinated.

When a novel file is created, a replication policy is used via the cues as well as and therefore it contacts the configuration service to run into if a while inwards the tabular array matches the policy. If no while matches the policy, the asking is forwarded to a random server matching that policy. In addition, WheelFS uses write-local policy inwards which the primary of a newly created file is the local server yesteryear default. This policy enables faster writes.

For replication, WheelFS uses primary/backup replication. For each while at that spot is a primary server as well as to a greater extent than or less backup servers. However, this scheme causes 2 problems: Firstly, since all operations transcend through the primary as well as updates require the primary to aspect ACKs from all backups; this replication scheme may campaign pregnant delays inwards wide-area systems. Secondly, if the .SyncLevel cue is used the replicas may locomote to a greater extent than or less of the updates also late. So, if the primary dies, a backup which replaces the primary may missy to a greater extent than or less updates as well as needs to acquire the missing updates from other backups.

By default, WheelFS uses close-to-open consistency. But inwards instance of a primary failure, all operations volition pick out to delay waiting for the novel primary to start. To avoid this delay, WheelFS provides .EventualConsistency cue that tin endure used whenever consistency requirements are non strict. In addition, WheelFS uses a write-through cache that improves consistency yesteryear writing the copies of each updates inwards cache to disk amongst the terms of increased latency.

When clients necessitate to locomote the information inwards their cache, they necessitate to locomote an object lease from the primary to save consistency. But this also brings additional latency terms since the primary needs to aspect for all leases to consummate to brand an update on the object.

In the experiments, they acquaint a disclose of applications that tin endure built on altitude of WheelFS such equally distributed spider web cache, electronic mail service as well as file distribution service. Distributed spider web cache application shows that it provides a comparable performance to pop systems such equally CoralCDN. In addition, inwards instance of failures, if eventual consistency is used, it provides consistently high throughput. In file distribution experiment, they revealed that amongst the handle of locality provided via large read cues, it achieves faster file distribution than BitTorrent. Finally, comparing of WheelFS to NFSv4 shows that it is to a greater extent than scalable thank you lot to the distributed caching mechanism.

Share this post