erd@romeo.cs.duke.edu (Ed Darken) (02/04/89)
Hi. I have a disk transfer rate question for all you Mac II experts out there. I wrote a MPW 2.0.2 tool called "disktime," in C, which allocates a linked list of buffers and then writes those buffers to disk using "write" (i.e., unbuffered I/O), keeping track of how long the writing takes. I ran this program on a Mac II with a Jasmine ID 90, with system version 6.0.2 and finder 6.1. The program takes two arguments: the number of 512-byte buffers to create and write, and the name of the file to write to. The only part of the program that gets timed is the loop that goes through the linked list and writes buffers to disk. Here are some of disktime's initial results: Time for 64 512-byte blocks: 0.13 secs. = 240.00 Kb/sec Time for 128 512-byte blocks: 0.28 secs. = 225.88 Kb/sec Time for 256 512-byte blocks: 0.63 secs. = 202.11 Kb/sec Time for 512 512-byte blocks: 9.78 secs. = 26.17 Kb/sec I then ran the Symantec Utilities' HDTuneup, which unfragmented the disk. Here are some examples I then got from disktime: Time for 64 512-byte blocks: 0.08 secs. = 384.00 Kb/sec Time for 128 512-byte blocks: 0.18 secs. = 349.09 Kb/sec Time for 256 512-byte blocks: 0.37 secs. = 349.09 Kb/sec Time for 512 512-byte blocks: 9.08 secs. = 28.18 Kb/sec I can tell myself a believable story about unfragmented files to explain the improvement in the second set of numbers. Can anybody explain the order of magnitude change in the transfer rate after the number of buffers surpasses some number between 256 and 512? At first I surmised the disk controller might have a RAM buffer of its own, so things would slow down when the buffer got filled. But if that's the case, I wouldn't expect to see any improvement in the 64-128-256 buffer tests after unfragmenting the disk. I don't understand this. Thanks, Ed Darken erd@cs.duke.edu
alexis@ccnysci.UUCP (Alexis Rosen) (02/24/89)
A few weeks ago, Ed Darken (erd@cs.duke.edu) posted a note describing an experiment he did benchmarking disk transfer speeds on the Macintosh. He described the unusual (and disturbing) result that when he transfered up to 256 blocks, he got results in the 350-380 KBytes per second, but transfers of 512 blocks displayed speeds an order of magnitude slower (only *28* KBytes per second). Has anyone figured out what cuases this? Was the experiment faulty, or has Ed uncovered a serious flaw in the File System, or is it some other possibility that I haven't thought of? (He defragmented his disk, so that's not it.) One thing I do know it that it's not the hard disk he used- those Wrens can take data considerably faster than the Mac can spit it out. Alexis Rosen alexis@ccnysci.uucp