The Kosmos distributed FS
First, the library buffers the incoming write operations and waits for the cache memory reserved for this purpose to fill or for the application to issue a flush command before pushing the data to the chunk servers.
Immediately after the data arrive, they become available for further operations.
Besides the outgoing data, the client library also buffers any metadata that are requested for 30 seconds. This helps to avoid unnecessary, repeated server contact.
If a client is running on a chunk server, it retrieves the data locally rather than using up network bandwidth. If a chunk server suddenly fails during a read operation, the client library automatically switches to another chunk server. All of this is completely transparent for the application.
Kosmos FS is an interesting alternative to HDFS and Google FS, but it is still at an early stage of development. Currently, one weak point is the metaservers. They need to be able to deliver metadata quickly. After all, to be able to process the file, a client needs to know which node the file it requires is stored on. If the metaserver fails completely, the files on the chunk servers it manages are also unreachable.
Because the metaserver additionally handles load distribution, it is responsible for the performance of the KFS network it manages. Unfortunately, there is currently no replication plan for metadata, in contrast to the scheme used by the chunk servers. Administrators need to take care of this manually and back up the data regularly.
Another issue is the lack of access controls. Currently, users can store any data on the distributed filesystem and read any data stored there. For this reason, KFS should only be deployed in trusted environments until a more mature version is released.
Buy this article as PDF
Competitors get in the game with RHEL without Red Hat
Security researchers have already notified Microsoft; some fixes are available
The company is collaborating with Google and Intel to use Kubernetes as an engine for Fuel
Customers can take a free test drive of SLES for HPC on the Azure Cloud
San Francisco-based chip company announces their first fully open source chip platform.
The whole distro gets rebuilt on glibc 2.3
Ubuntu Vendor tries to solve app packaging and distribution problem across distributions.
Founder of ownCloud launches the Nextcloud project.
Will The Machine change the way future programmers think about memory?