OpenAFS logo
NJIT

Wednesday 21 May 9:00am

Location: Second Floor of Campus Center at NJIT

Speaker: Alistair Ferguson (Morgan Stanley)

Title: OpenAFS and the Dawn of a New Era

Abstract:

AFS was introduced into Morgan Stanley around 1995 as part of the Aurora project (see LISA'95 paper).  Aurora aimed to allow tens of thousands of machines distributed across the globe to be centrally managed.  At the time, AFS had a number of significant advantages over similar products (e.g. NFS) including a single hierarchy, efficient local caching, guaranteed cache consistency, read-only data replication, seamless data migration and security (using ACL's and Kerberos authentication).  These features allowed a relatively large (at the time) number of machines per AFS cell, with homogeneity guaranteed by the fact that even the OS was in read-only (RO) AFS.  In 1995, there were around 26 cells each with around 20GB of replicated data.  There was also around 200GB read-write (RW) data.  Aurora was remarkably successful. The number of machines has grown from a few hundred in 1996 to almost 22,000 at the end of 2007.  Similarly, the AFS plant has grown to over 60 RO cells, most serving almost 3TB of data.  As predicted in 1995, the RW data has grown explosively, rising to over 14TB served from 6 dedicated RW AFS cells.

In that time, the computing environment has changed significantly.  Aurora was designed when disks either had little capacity or were very expensive, now they have large capacities and are cheap.  Network capacity has also increased significantly.  So for OpenAFS, is there any reason to be optimistic ?

The history of AFS at Morgan Stanley will be covered, including the original design goals and some of the constraints involved. Those constraints will be re-evaluated in the light of modern technology, as well as changing requirements, notably the need to support hundreds of thousands of heterogeneous machines and reduce costs.  That naturally gives rise to a discussion of the advantages of OpenAFS in an enterprise, and the role it can play in a complex, fast moving, highly critical production environment.  Finally, some of the recently completed projects and the major impact these have had for Morgan Stanley will be highlighted, as well as possible future projects / directions.

Biography:

Alistair Ferguson received a B.Sc. in Computer Science from Teesside Polytechnic in 1991, before doing a Ph.D. in Artificial Neural Networks at the University of Hertfordshire in 1995.  After becoming a research fellow and getting married, he left the university in 1997 to earn some money.  After a year consulting, he joined Morgan Stanley as a Unix System Administrator in 1998 and has been involved in AFS ever since. Starting with operational responsibilities for AFS, he became the lead engineer responsible for OpenAFS in the middle of 2007.  He can be contacted at alistair.ferguson at morganstanley.com.

Slides: PDF

AFS & Kerberos Best Practices Workshop 2008: Wednesday Keynote