timk@NCSA.UIUC.EDU (Tim Krauskopf) (08/06/89)
Previous discussions of Mac FTP transfer speeds have let the Ethernet boards off the hook, but ask the question: why does copying a file time at approx. 150KB/sec and NCSA Telnet time at 50KB/sec with MacTCP, 30KB/sec without MacTCP? I admit to some software overhead in NCSA Telnet, but have you ever done the following experiment? Try disk read sizes of 100B,1K,8K,32K,64K,1024K and see what the data rates are. Finder copy uses very large buffers now, usually as much system memory as it can get its hands on. The larger the block, the faster it goes. Track sizes are getting to be 32-64K range -- remember the Mac II uses one-to-one interleave, and seek times are still significant overhead. NCSA Telnet blocks the network 4K (TCP window) at a time, but it does 8K at a time to disk to save memory. Try recompiling it with 64K blocking to disk and let me know how much it improved. For some reason the system disk cache has enough overhead so that it is not a cure-all for this problem. Tim Krauskopf timk@ncsa.uiuc.edu (ARPA) National Center for Supercomputing Applications (NCSA) University of Illinois, Urbana-Champaign
amanda@intercon.uu.net (Amanda Walker) (08/07/89)
In article <8908052109.AA12473@zaphod.ncsa.uiuc.edu>, timk@NCSA.UIUC.EDU (Tim Krauskopf) writes: > NCSA Telnet blocks the network 4K (TCP window) at a time, but it does > 8K at a time to disk to save memory. Try recompiling it with 64K blocking > to disk and let me know how much it improved. Actually, it occured to me this weekend that the problem may be a little more subtle. Telnet writes out files by using FSWrite--I remember a discussion a while back about FSWrite, where someone claimed that FSWrite just calls PBWrite 512 bytes at a time. I haven't gotten around to changing the code to just call PBWrite directly, but this might make the file system a lot happier... -- Amanda Walker InterCon Systems Corporation -- amanda@intercon.uu.net | ...!uunet!intercon!amanda