SGI's NFS isn't really crappy at all, its actually supposed to be
the best in the biz. The problem is really NFS abuse. You have to
remember that 10BaseT is only good for about 750K/Second which is
exceedingly slow with repect to even the slowest of direct SCSI
disks. Then there's the client-server overhead of NFS on top of that.
No database or high performance website should ever be NFS mounted.
Even if you use 100Mbit FDDI, its still only 25% as fast as a 4O
MByte/sec Ultra-Wide SCSI drive.
Databases like Texis/Webinator add another dimension to the problem;
They use shared memory to control concurrency access to the database
when there are multiple users reading and writing to the RDBMS.
If the RDBMS is NFS mounted and users on both machines attempt access
simultaneously, the database may be trashed.
This is just the way it goes. HTTPD's do not provide web-walkers with enough
information to allow them to avoid completely re-walking a site when looking
for new/changed documents. The HEAD operation only says that a parent document
has been changed, it says nothing about its children.
Lets say we have a web server that has a tree based hyperlink structure with
9 documents in it.
H
1 2 3 4
5 6 7 8<---changed
The GW program will have no way of knowing from document 2 that documents
5 and 6 remain unchanged, and examination of document 4 will not actually
give information about the changes to 8.
This oversimplifies the problem too. Most servers look like directed graphs
and not trees.
Yes, see -x in http://www.thunderstone.com/gwman/node21.html
Nope, It considers alt text to be a valid part of the document content.
This is because graphics often contain useful textual content which
is of course only readable through the author's ALT text comment.
Thunderstone