9fans archive / 1997 / 12 / 25 / prev next From: G. David Butler gdb@dbS... Subject: [9fans] Re: Kfs file server Date: Thu, 11 Dec 1997 08:10:16 -0600 From: Kenji Arisawa <arisawa@aic...> >Kfs file server of plan9 looks too slow. [snip] >Coping on 9fs involves bi-directional transmission >through 10 base ethernet using 9P protocol. >The value is reasonable and may be efficient. >If we apply 100 base ethernet, this time will largely improved. What are you using as a file server? A plan9 fileserver works very well. >However why the result of kfs is so bad? >Caching problem? This is the main reason. The kfs system does no caching of the hard drive. There have been other discussions of this topic and the best idea forwarded so far is to integrate the mount driver with the memory system to use "unused" memory for a cache. This would improve both kfs and 9p. Using the QID of the file to flush old contents keeps the cache in sync with the fileserver. This implies that each block transfered from the cache should wait for a walk from the fileserver to verify the QID. Doing it just on the open is not good enough. Of course, the cache has to be write-through. >Algorithm? It is the same one in the fileserver. It is simple and fast. >Can I improve this value? Easily. David Butler gdb@dbS...