Lustre was developed as part of a DoE funded project that started in about
2002. You can find all sorts of technical information at
http://www.lustre.org/index.html . The creator of Lustre, Peter Braam, came
from CMU and Intermezzo, Coda, and all that. However, even though Lustre may
share some resemblance to those other file systems, this is due to the
common thread of Peter Braam carrying it does not necessarily share a code
base with any of them.
GFS was developed at the University of Minnesota and then by Sistina which
was a startup that grew out of the UofMN. Sistina was later acquired by
Redhat where GFS currently resides.
Lustre and GFS both appear as file systems that run in a clustered computer
environment giving all clients in the cluster a standard POSIX file system
semantics (open, close, read, write, fcntl, .etc.) as well as concurrent
read and write access to files. From the application point of view, they
essentially look the same. The difference comes in when one considers the
fundamental access method that Lustre and GFS use to access data. GFS uses
traditional block-based access protocols (i.e. SCSI) to storage devices on a
SAN where the storage devices are traditional block-based devices (i.e. the
read and write sectors on disks). Lustre assumes that the storage devices
are *intelligent* storage devices (a.k.a. loosely defined Object-based
Storage Devices or OBSDs which is NOT to be confused with object-oriented
devices which don''t exist and let''s hope they never do) and
use a more
advanced object-based protocol to access data on the storage devices. There
are traditional block-based storage devices at the end of the protocol chain
where the actual bits get stored but the thing that is important here is to
understand that Lustre does not talk directly to the block-based storage
devices - it talks to object-based storage devices also know as Object
Storage Targets or OST''s.
Both GFS and Lustre probably do an equivalent job at managing files in a
relatively smallish file system of several terabytes with a smallish number
of clients - something under 32 or so. The big difference comes in when
scaling the number of clients or the size of the file system or both to very
large numbers - like hundreds, thousands, ten of thousands, or hundreds of
thousands of clients on one file system that is in the Petabyte to Exabyte
range with trillions of files. The only way to provide scalability like this
is to use the Lustre approach of using *intelligent* storage devices and
clustered metadata servers and an associated object-based protocol.
Block-based file systems cannot scale in these dimensions to these extremes.
If you don''t believe me, try it.
Bottom line is that GFS is a good choice for smallish clusters and Lustre is
a good choice to medium to ridiculously large clusters although it will work
at any size.
In terms of reliability, availability, serviceability, and all those
*enterprise-class* ilities - you need to talk to the Lustre/GFS marketing
folks about that: Welcome to Marketing - 2 Drink Minimum.
Does that help?
-Tom
_____
From: lustre-discuss-bounces@clusterfs.com
[mailto:lustre-discuss-bounces@clusterfs.com] On Behalf Of Screaming Eagle
Sent: Sunday, February 19, 2006 8:13 AM
To: lustre-discuss@clusterfs.com
Subject: [Lustre-discuss] Lustre vs GFS...
All,
I am a newbie to this. Is Lustre an offshoot of GFS? What are pro''s and
con
btw Lustre and GFS.
Thanks.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20060219/9ee44eb4/attachment.html