Lustre Distributed File System Component - Architecture

Lustre File system Component
1)MDS ( Meta Data Server ) & MDT (Meta Data Target)
2)Oss (object Storage server ) & (Object Storage Target)
3) Client
4) Lustre Network (LNET)
Lustre File system Capacity depends on Lustre file system Object Storage Server.

1)Lustre File system Implementation.
Linux driver module is loaded into kernel and file system is mounted like any other file system.
1.1) OBJECT STORAGE TARGET
OST stores file data as data object.

2)DATA OBJECT AND FILE STRIPING
in ext3 file system inodes are point to data block.Lustre Distributed file system inodes are not point to data blocks instead of point to one or more data object.

NOTE:number of object per file is user configurable & it can be tune to optimize performance.

3) When CLIENT OPEN A FILE.
a)For read & write operation client interprets layout of LOV(Logical object volume) layer in MDS,which maps  offset and size one ore more object residing on separate OST.
b)FILE OPEN OPERATION it transfer the set of object pointer from MDS to the client.So that client can directly interact with OSS allowing client to perform I/O on the file without further Communication on MDS.
c)and client  locks the file range being operate on and execute one or more parallel read or write operation directly to OST.
d)By through these method  bottleneck  for client-TO-OST communication are eliminate.

NOTE :MDS Contain  filename,directories, permission and file layout.layout contains OST number, Object Identifier & Points to one or more objects associated with data file.
In the Other Words
a) The lustre client obtain the file information from MDS.
b) Once this information is obtained, then client interact directly to OST on which file is Stripped.

4) PURPOSE FOR STRIPING IN LUSTRE FILE SYSTEM.
Stripping contains the on of objects.Number of object is called stripe_count.

5) HIGH AVAILABILITY
a)Lustre
MDS are configure in Active/Passive mode [ We can't maintain the Lustre mds as active/acitve because if we maintain then two mds create the duplication - Only one mds is possible.]
OSS are configure in Active/Active mode [If we maintain the extra oss server then it will reduce the work load of the server and act as load balancing ]Provide without extra overhead.

6) Lustre Network,
lctl -Lustre Control via ioctl interface ioctl - Input Output Control.
/etc/modprobe.conf file we are adding the module parameter.


Lustre Distributed File system Measurement by Following Parameter
1)Distributed File system Total Size
2)Data Transfer Bandwidth.
3)Supporting No of clients.
Standard Benchmark is IOR Benchmark

Related post