djc@duke.cs.duke.edu (David J. Cherveny) (03/24/91)
Could anyone comment on the merits/problems with using ASYNC NFS. The write performance speedup is phenomenal but I worry about integrity after a crash. I know its not recomended for databases, etc. Also, could several workstations overwhelm a net and/or server? David Cherveny djc@rm2000.mc.duke.edu djc@cs.duke.edu
ash@hpindda.cup.hp.com (Art Harkin) (03/26/91)
/ comp.sys.hp / djc@duke.cs.duke.edu (David J. Cherveny) / Mar 24, 1991 / Could anyone comment on the merits/problems with using ASYNC NFS. The write performance speedup is phenomenal but I worry about integrity after a crash. I know its not recommended for databases, etc. Also, could several workstations overwhelm a net and/or server? David Cherveny djc@rm2000.mc.duke.edu djc@cs.duke.edu ---------- My understanding of the "-async" option is that it just allows the NFS client not to wait for acknowledgments that data has been written from the disk buffer to the disk on the NFS server. The reliability is somewhat decreased by NFS standards, but only to the level of what a user would see if he was accessing the disk directly on the server itself. In other words, NFS adds reliability to the normal disk buffers on the server by sending the data again after a data loss (a crash), where as local programs on the server would not even have been guaranteed this reliability. So for most users, there is no reason not to use this option, other then the ones you already stated which is the synchronized write (via ioctl method) employed by some data integrity conscious database programs. This is my understanding and there may be some HP NFS experts who could correct or add to my comments. Art Harkin ash@hpda.cup.hp.com Hewlett Packard Cupertino, CA