Back

* I added multiple AFP enumeration caches (16 of them) so when multiple directories are simultaneously being enumerated with GetFInfo or GetCatInfo, those enumerations were not throwing out each other's data. This was the bug I was supposed to fix.

* I noticed the enumeration caches were small (512 bytes) -- the same size they'd been since 1987 when memory was very tight on a Mac. I increased size of those enumeration caches. The result is enumerating larger directories is done with fewer transactions with the server.

2/5

* I made MakeFSSpec much faster. When System 7.0 first shipped, I got a developer bug that showed MakeFSSpec (new in System 7) was very slow on AppleShare. I filed a bug for the developer and even explained what the problem was. The code would enumerate a directory to find a matching filename to the filename passed to MakeFSSpec. If the filename was the last to be seen in a directory enumeration, that resulted in many transactions. With my rewrite, making an FSSpec is accomplished with only one or two transactions with the server. This bug effected creating FSRefs, too.

* Increased size of response buffers used for ATP to hold 8 full-sized ATP response packets (it was 1) to increase network performance (reduce the number of transactions with the server). Why was it so small? The code had not changed since 1987.

3/5

* I replaced routines that moved blocks of memory 1-byte at a time in loop with calls to BlockMoveData for improved performance. Someone at sometime thought BlockMove/BlockMoveData couldn't be called at interrupt time and added that slow memory copy routine -- they were wrong.

* I replaced BlockMove calls with BlockMoveData wherever possible (BlockMove was only needed when moving code) for improved performance.

* I locked out loading on Mac OS versions earlier than 7.6 (this version of AppleShare was only tested with Mac OS 9.1/9.2.x but it should work back to 7.6). That's because code in the AppleShare client had required features only in 7.6 and later.

* I removed dead pre-System 7.6 code (much of it still ran during initialization but did nothing).

4/5

* I replaced multiple INIT resource with a single INIT resource that loads the other code (now in CUST resources) in the correct order. If the AppleShare client was edited with a resource editor on Mac OS 9.1 or later, those INIT resource might have loaded in the wrong order.

* Fixed a few other random bugs I saw that probably wouldn't be noticed by anyone else but they bugged me.

5/5 end of

From the 9.2.2 CD

@atax1a It was! I was in a very small group tasked with improving performance and stability from 7.5.5 through 9.2.2. We fixed bugs wherever they popped up if there was nobody already working on that code (mostly non-UI code). We also helped other teams find and fix those kinds of problems.

@jimluther thanks for your work and for sharing your stories :)

@theirongiant @atax1a At the time, I was measuring performance gains on local AppleShare servers, and those gains were great.

Those changes to enumeration and creating FSSpecs/Refs make reading directories much faster -- especially when multiple directories are being read at the same time which is what happens if you copy of directory that contains subdirectories.

Those changes and the ATP buffer changes also reduce the number of network transactions. Over the internet, that makes a HUGE difference because of network latency (ping). Here’s an example. By making the ATP buffers 8x larger, there are up to 8x fewer transactions when copying a file. If the connection to a server has 100ms ping… well, you can do the math to see that reducing transactions can remove LOTS of time.

@dgentry Nope. The Blue Meanies group was around from before System 7.0 to maybe just before 7.5. I was in the Developer Technical Support during that time, and joined the System Update team (which became Mac OS Release Engineering after Copland was cancelled) just as 7.5.3 was being wrapped up.

@pierrenick Apple used to pay much more attention to backwards compatibility. However, getting the testing resources to test changes on older systems is hard, so that's why it isn't done as much as everyone would like.

The AppleShare extension code I changed had been tested with systems back to 7.6 (at least I was told that) and the changes I made to the AppleShare extension did not depend on anything new, so I feel fairly safe saying the AppleShare extension v 3.9.2 will work with Mac OS 7.6.

@atlauren Anyway, after I quit working on webdavfs, it never really got any faster but we came up with a way for webdavfs to communicate with the Finder to show progress during close on the PUT request to the server. Bill Conway, who had taken over responsibility for webdavfs and implemented that change in webdavfs. The Finder team showed that progress during close to let the user know something was happening during file copies.

Sometimes showing something is happening in UI makes users *think* things are faster

2/2