Key aspects of data storage
- Be able to share our work across a network
- $ expensive storage
- reduce the need of local storage
- need to keep data safe
Unix File System
File Replication requirement
HW and SO heterogeneity requirement
Fault Tolerant requirement
File Service Architecture
Quick comparison DFS and Unix FS
The NFS File System
- fault tolerant
Not fully supported
NFS Architecture details
- Require a kernel module to add support to nfs
- Shared cache of R.U. Blocks
- Reduce the request sent to the server by caching
- Access Control enhanced thru KerberOS and DES
- Same approach that Unix thru know techniques: read-ahead, delayed write
- client: On updates: extra measures needed: write-through or using commit
- client: (T - Tc < t) v (Tm(client) = Tm(server))
mount service details
- modification to mount
- hard mount / soft mount
- automount on demand
The AFS File System
- Research project from Carnegie Mellow University and IBM.
- Access thru normal Unix file primitives.
- Scalability is the main goal, caching the whole file in the client node.
- Two main process venus (client) and vice (server).
- Kernel with a small modification to capture open, update operations.
AFS design assumptions
- Small files < 10kb.
- Read operations are more common than write operations.
- Sequential access is common, random access rare.
- Most read/write operations by only one user, on shared files only one user writes.
- if a file has been referenced recently, it is usually probable that will be referenced in a short while.
Other DFS solutions
- Serverless network file system architecture.
- Distribute responsabilities across a set of available computers.
- software RAID .
- comparable solution to NFS and AFS with better overall performance
- Fast switched LANs available.
- increased demand for access to shared data.
- limitations of systems based on central file servers.
New DFS solutions
GFS (Google File System)
- Developed by Google© to meet demands of Google's data processing needs.
- Components failure are the norm rather than the exception.
- built with storage machines using inexpensive commodity parts.
- Files are huge by traditional standard.
- Files are mutated by appending new data, rather than overwriting.
- Requires specific modification to apps (it is non POSIX).
- A GFS cluster consists of a single master, multiple chunkservers and
New DFS solutions
- Scaling to several petabytes (1 thousand Tera bytes)
- High scalability and performance in mind
- RDMA or TCP/IP Interconnect.
- Stackable user space design.
- uses the FUSE facility as a bridge between user space and kernel
- Offers native NFS support.
Network Attached Storages (NAS)
Use a spacebar or arrow keys to navigate