Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!cs.utexas.edu!uunet!munnari.oz.au!uhccux!bmartin From: bmartin@uhccux.uhcc.Hawaii.Edu (Brian Martin) Newsgroups: comp.sys.mac.apps Subject: Re: 4th Dimension & Networking over the phone Message-ID: <10559@uhccux.uhcc.Hawaii.Edu> Date: 9 Dec 90 03:37:15 GMT References: <1990Nov23.000340.1514@gpu.utcs.utoronto.ca> <10381@uhccux.uhcc.Hawaii.Edu> Reply-To: martin@medix.pegasus.com Organization: University of Hawaii Lines: 57 In article urlichs@smurf.sub.org (Matthias Urlichs) writes: >In comp.sys.mac.comm, article <10381@uhccux.uhcc.Hawaii.Edu>, > bmartin@uhccux.UUCP (Brian Martin) writes: >< In article <1990Nov23.000340.1514@gpu.utcs.utoronto.ca> judge@gpu.utcs.utoronto.ca (Peter Judge) writes: > >< Alternatively, you could run "headless" macs at your central >< site, each with timbuktu/remote or carbon copy mac running, and access >< these macs via v.42bis modems from your remote sites. That may >< be a reasonable implementation, since you transfer screen images rather than >< data across the phone lines. You could enter and review data on-screen, but >< you wouldn't be able to print at a remote site using this approach. >< >Also, screen access should be significantly slower than shipping the raw data >over the line. You'd also tie up one mac per remote user, which is >prohibitively expensive. Here's why it doesn't make sense to ship raw data over a line. Let's say you have a 4D database with 70,000 records in one of its files (I have one). Let's also assume that each record contains, on average, 100 bytes. Guess what happens when you want to do a simple search, such as: SEARCH([Patients];[Patient]ZipCode="96786") 4D downloads all 70,000 records (about 7,000,000 bytes) from the fileserver to your machine, and does the entire search on your machine. Guess how long it takes to download 7MB, even at 38,000 baud? Compare that to 1,000,000 baud download speed via Ethernet to a local Mac running 4D. You could optimize this by indexing the field you plan to search on. In that case, with overhead, you're still going to download 700,000 bytes (10 bytes of data & index overhead per indexed record * 70,000 records) just to do a simple indexed search. That's still a lot of overhead for phone lines. I've actually run all these tests, and we currently use timbuktu/remote with headless macs for dialup access, as it provides the _only_ reasonable performance. Although this ties up one mac per dialup line, the development and maintenance costs may be cheaper than going to a conventional client server model, where the data server would be Sybase, Ingres or Oracle. Using the Mac IILC, the cost per dialup line would probably be less than $3,500, including two v.42bis modems and two copies of Timbuktu Remote. Best regards, Brian K. Martin, M.D. 1103 9th Avenue, Suite 203 Honolulu, Hawai`i 96816-2403 Voice (808) 733-2003 Fax (808) 733-2011 INTERNET: martin@medix.pegasus.com, bmartin@uhccux.uhcc.hawaii.edu ARPA: uhccux!bmartin@nosc.MIL