|Levanta Releases Linux Management Code|
October 31, 2005
Linux system management software vendor Levanta Monday released its MapFS Linux file system management code to the open source community.
MapFS, a key component in San Mateo, Calif.-based Levanta's Linux management appliance, is a virtual file system that simplifies data sharing between multiple Linux machines connected to a shared storage medium, the company said.
MapFS is a Linux kernel-loadable module. It was developed under the GNU GPL (General Public License) in early 2004 and has been shipping with Levanta front-line product, Intrepid M, since last year.
"Linux management is one of the hottest development areas in open source today," said Levanta CEO Matt Mosman.
"By nature, Linux is extremely well-suited for advanced techniques in server provisioning, disaster recovery, change management and other common systems management scenarios where, frankly, Windows management solutions have been relatively stagnant.
"Linux is where the management innovation is happening today, and by releasing MapFS, Levanta is introducing a mature, free code base that introduces interesting new virtualization techniques that we believe the development community will find compelling."
MapFS enables a Linux file system to utilize copy-on-write functionality. It also enables component file systems to be combined into a single virtual file system that is fully writable.
Product manager David Dennis told Ziff Davis Internet Monday that although there is much development currently happening in virtualization at companies such as Zen, VMWare and Globus, "we did not see a lot of activity involving file systems. The reason we released this now is because our technology is mature and stable enough to stand up in any OS stack.
"We also believe that there are people (in the open source community) who will come up with some enhancements we haven't thought of, and that'll make it that much better," Dennis said.
With MapFS, a Linux system can provide sharing of read-only file systems while at the same time providing each client of the read-only file system the ability to write to its own data store, Dennis said.
Files can be either on a read-only persistent repository file system, or on a writable persistent overlay file system, he added.
This enables an "optimistic sharing," where everything on the file system is assumed to be read-only.
If an attempt is made to modify a file—that is, a private copy is needed—the performance hit is typically minimal, because most written-to files are small. Even in the event of a larger file, the performance hit is a one-time cost, Dennis said.
How important is this release to the development community at large?
"This is hard to estimate based on what appears in the announcement," senior analyst Mike Karp of Enterprise Management Associates in Westboro, Mass. told Ziff Davis Internet.
"Its significance will be a function of how much the open source community sees this as differentiated from what is available on the Red Hat, SUSE and other distributions."
Other versions of Linux file virtualization are available from Cluster File Systems (Lustre), Hewlett-Packard Co. (which uses Lustre but call it SFS), IBM and ADIC, among others, Karp said via email.
"All the focus is on I/O and filing systems for high-performance computing environments, which invariably means 64-bit," he added.
"Any form of storage virtualization offers tremendous opportunities for IT managers who want the ability to manage large storage installations as if all components with the installation were part of a common pool," Karp said.
"Virtualized file systems offer the advantage of virtualizing data at the file level, allowing storage management to optimize the way the data itself is accessed. The idea of a read-only system is an interesting approach to this sort of management, and would seem to eliminate the need for the overhead typically associated with file locking and the need to upload changed data to the server," Karp said.