Hi Jos,

Well I haven't done the reading you have but it is still my take that leaving it operational is unwise unless the hardware/driver(s) support write completion on power-up.
Even then I have my doubts.
Several years ago now, in a Netware 3.11 environment and using a Storageworks RAID array I argued strenuously with the network Admin. to turn on write caching. She refused. I escalated. She still refuse. She won. And I later came to feel, and told her, that I was glad she stuck to her guns (after hearing of other misadventures related to it).

More recently it was unknowingly on on my workstation, operating in a simple peer-to-peer setup. I got an error during a Word save (well sevceral seconds later, if I remember, about 'delayed write buffer no longer available... data lost' even though the save was to the same machine (i.e. the network was not implicated. I also saw the message once relating to a VFP table that was using the network.
It's been 2+ years now since I turned it off and I have not seen such messages since here. I was at a client's 2 weeks ago and that message appeared there on his WinXP-Pro system. A quick check revealed that his write behind cache was on. We turned it off. Though he continues to use the same application, and even more efficiently/quickly than earlier, no more 'data lost' message!

In summary, I think there is still a gap in logic between the OS and the hardware and that sometimes the OS gets confused when write behind caching is on. So I feel it is better left off regardless.

By the way, the 'bible' on SQL Server (V7) states emphatically that the performance hit of turning it off is negligible for that application.

cheers


>Hi Jim. I have done some web reading on various forums on this topic and it seems a bit unclear as to what the best choice iro switching this feature on or off is.
>
>On the one hand switching it off allows for more security in getting the data written to disk but on the other there are posts that argue that forcing writes which could be delayed can have a performance impact if the server is also performing a lot of reads.
>
>Then others comment that on todays hdd's the performance might be neglible and also that it depends on the purpose of the server. For instance, if the server is a web server with little disk writing then leaving it on can be better than off, especially if the server is protected with a ups.
>
>It seems to me that the question of whether to leave it on or off is server function dependent.
>
>
>>I'm not the famous DougH but I thought I'd offer my 2-cents worth...
>>
>>Yes, write caching (aka write behind cahcing) should ALWAYS be turned OFF in any HD setup, even RAID, UNLESS it has a controller specifically designed and capable, with appropriate intelligence in its controller/driver/etc to correctly handle a sudden power loss situation. Such controllers are very rare at this time, though they do exist and cost significant extra $$$ I believe.
>>
>>I would bet that the write caching at the file servers is already turned off. Not that it would have come setup that way, but rather that people who maintain these things for a living know this to be the case.
>>
>>My *personal* opinion is that write caching on a local drive can still cause problems, even during regular processing and without a sudden power failure. I had two illogical experiences with errors involving the cache (one with Word and one with a VFP record) and neither has recurred in about 2 years since turning it off.
>>
>>good luck
>>
>>