On Thu, 9 Mar 2006, Alan Brown wrote: > On Thu, 9 Mar 2006, Hugh Sasse wrote: > > > > Also enable server mode. enable caching of temp dir > > > > Enabling server mode is not possible given some types of client (from > > my reading of the docs). I don't know all the clients people use, so > > I must err on the side of caution. Or is that paranoid in 2006? > > Server mode is only risky under the following circumstrances: > > 1: Access is made via POP3 and local disk (or a non-compatible imap3 server) > which use non-compatible locking methods. > > 2: Clients _simultaneously_ access using both methods. > > I used server mode for many years in a mixed environment with no > problemss - by ensuring the lock mechanisms used were compatible. > > Even without that, it is highly unusual for users to use both pop3 and > local disk access methods - those that do are usually technically savvy > enough to understand the corruption risk and not use both methods > simultaneously once it is explained to them.
OK, then I will certainly enable it for the large users. It's just that usage patterns are a "known unknown", and I'm reluctant to get caught by the "unknown unknowns" :-) > > > Thanks. I'm not au fait with disk internals, and thought that some > > disks may have many heads > > They do (one per platter), but.... > > >, not just to read one cylinder at a time, > > Only one head is ever active at one time with current commonly available > commercial available disk techmology. > > The only way to achieve what you want is to use a suitable hardware > controller capable of simultaneously addressing multiple drive busses > (despite the other advantages of scsi, only one drive can be addressed > at a time on any given physical scsi bus. OK, I thought there might have been developments I'd not kept up with. Thanks. > > Large scale (S)ATA or SASI raid controllers give better results most of > the time in terms of overall bandwidth. Latencies cannot be redcued > below seek+rotation times even with fancy controllers. > > For ultimate bandwidth andlatencies, suitable solid state arrays are the > only way to go - but are extremely expensive. To be nonvolatile and fast they would be. I'm trying to get another disk for this system, but I doubt they'd stretch to RAID. Thank you anyway, it's useful stuff to keep in mind. > Hugh