I'm interested in knowing why having a lot of objects cached is important? I'm especially interested in the expected cache hit rate for your anticipated object store?

Normal internet object request distribution shows a ziff-like curve - meaning if you assign the most requested object a score of 100 - the next most hit object will be assigned a score of 50 - and so on. (it's not always exact ziff - but in general it follows the same curves)

By the time you reach object 1/2 million - os so - the request rate is so low that the chances are very high that the object will have expired, or have been modified, or deleted.

Do you see this same behavior in your environment? If not why? I'm really interested.

Perhaps you need a replicated object store - and not a caching object store - a different beast.

thanks,

dp

Chris Good wrote:

Nick Hilliard wrote:

...

Perhaps this is o/s revision dependent, but 4.x can have many millions of objects.

Bzzzzt. 256bytes RAM per object so on a C760 only 4M objects. We were hoping to cache pure html (which would never be more than a few K in size) and 200M documents was the minimum number of documents for things to be usefull to us. [sigh]

-- Chris Good - Dialog Corp. The Westbrook Centre, Milton Rd, Cambridge UK Phone: 01223 715006 Mobile: 07801 788997 http://www.dialog.com

-- Get answers NOW! - NetApp On the Web - http://www.netapp.com http://now.netapp.com