Huge Loss of Available RAM

Brought to you by

Copyright 1997 by Bill Fox

-- 6/26/96 (rev.9, 9/20/97)

I. Problem (6/26/96-9/20/97):

After quitting an application that uses a very large amount of RAM (like Netscape) and Open Transport with TCP/IP version 1.1 or later with System 7.5.3 or later, the "About This Macintosh" (ATM)--now called "About This Computer" in MacOS 7.6, 7.6.1 and 8.0--under the Apple Menu reports that the Largest Unused Block (LUB) of available RAM is reduced by about the same amount as Netscape used when open. This memory loss or leak or fragmentation (whatever) can amount to 5-12 MB depending on the Netscape version and the number of pluggins and whether or not Virtual Memory is turned on. On computers with less than 24 MB of RAM this has been reported to cause a low memory error message during the subsequent use of applications that use a large amount of RAM like MS Word. This memory loss is reported by memory monitors other than ATM (e.g. RAM Gauge, SoBig and FreeMem) but not all (e.g. MacUser RAM Handler and Free RAM). One has to reboot to free the locked up RAM. The reported fixes of selecting "Load only when needed" in the TCP/IP configuration options and using MacOS Purge do not work.

II. Solution (8/29/96, rev.8 2/8/97, rev.9 9/20/97):

[9/20/97] From reader's comments and my own experience, this problem is back in Mac OS 8 using Open Transport 1.2 and OT/PPP 1.01 or just using many different applications. The problem covered here is related to internet usage only. However, in my experience the problem under Mac OS 8 is very occasional and difficult to reproduce consistently enough to isolate the problem.

[2/8/97] Update Open Transport and TCP/IP to version 1.1.1 or 1.1.2 and use Open Transport PPP 1.0 instead of FreePPP or MacPPP. OT 1.1.1 or 1.1.2 and OT/PPP1.0 may be obtained using the links on the my Software Update page of my site.

Experience Notes -- The memory loss problem was solved on my PCI Power Macintosh 8500/120, which uses a direct internet connection rather than a PPP connection, when I updated to version 1.1.1b6 of OT/TCP/IP on August 12th. OT 1.1.1b6 did nothing for the problem on my Nubus Power Macintosh 7100/80 and my Quadra 605 and both use a PPP modem connection with FreePPP 2.5v2. Rich Kubota suggested using OT/PPP 1.0f1c6, Apple's developing PPP utility for Open Transport, instead of FreePPP but this did not work. Updating to version OT 1.1.1b7 initially resolved the problem using FreePPP but I had to both quit Netscape and close the PPP connection to recover the memory loss -- even then this procedure very infrequently did not work and the memory loss was not recovered. Over time, the frequency of the memory loss increased to become permanent again. Then, I tried OT/PPP 1.0f1c6 instead of FreePPP and the memory loss was recovered. Alternately trying FreePPP and OT/PPP showed OT/PPP worked with OT 1.1.1b7 but FreePPP did not. I reinstalled FreePPP 2.5v2 again and the memory loss problem was again gone. After only two days of use, however, FreePPP "degraded" and the memory loss returned. Using OT/PPP1.0f1c6 again eliminated the memory loss. After several more days, it appears that together OT 1.1.1b7 and OT/PPP1.0f1c6 eliminate the memory loss problem most of the time. Occasionally, I still get the memory loss even after quitting both Netscape and OT/PPP1.0f1c6--I think it's related to the use of applications while the PPP connection is open but I am not sure. I almost never get the memory loss if I just run Netscape. This experience was the same for OT 1.1.1b8, the final version of OT 1.1.1 and OT 1.1.2 using the latest and final version OT/PPP 1.0.

You may either ignore the mega memory loss problem, if it causes you no other problems, or use the workaround solution below.

III. Workaround Solution (July 4, 1996):

First make a PPP connection with ConfigPPP or FreePPP, not Netscape. Then launch, run and quit a small application that uses TCP/IP and Open Transport. The smallest that I have found so far that accomplishes the task is Peter Lewis' Finger. I set the "User" to "Me" and the "Machine" to "1.2.3.4" and press the "Finger" button. In a few seconds I get an error response and I quit the application. I can then run and quit Netscape with locking up MB's of RAM. Netscape may then be run and quit many times until one has to reboot. Then the process has to be repeated. The ATM-reported loss is reduced to 100's of K rather than MB's.

The other solution is to revert to using MacTCP, at least on 68040 Macs experiencing the problem.

IV. Example (July 4, 1996):

PowerMac 7100/80

• System 7.5.3 Rev.2 using Open Transport with TCP/IP version 1.1
• 56 MB Physical RAM

_________________________________________________________________________

TABLE 1. -- RAM Loss with "Load only when needed" not checked in TCP/IP
_________________________________________________________________________

. . . . . . . . . . . . . . . . . . .Change (K) in Largest Unused Block of RAM
. . . . . . . . . . . . . . . . . . . .Reported by 'About This Macintosh'

1st App Using TCP/IP . . Virtual Memory Off . . Virtual Memory On

Netscape v2.02. . . . . . .-8794. . . . . . . . . . . . .-6232
Finger v1.5. . . . . . . . . . -793. . . . . . . . . . . . . .-458
OTTool. . . . . . . . . . . . -1284. . . . . . . . . . . . . .-722
WhatRoute v1.1. . . . . . -1445. . . . . . . . . . . . . .-780

_________________________________________________________________________

By turning Virtual Memory (VM) on, there is less of a loss. However, my computer then takes 41 seconds longer to boot and loses 57 MB of available hard drive space. I also experience other problems that I won't go into. You may wish to experiment with this but I don't need VM so I don't use it.

V. Interesting Observations (as of July 4, 1996):

1. Several memory monitoring utilities do not indicate a RAM loss, as noted above.

2. Reinstalling 7.5 Update 2 from the CD ROM eliminates the ATM-reported RAM loss for the first TCP/IP/OT use. When the computer is rebooted, the RAM loss problem comes back.

3. Tim Fredenburg suggested that I experiment by creating the memory loss and then trying to "use" the memory by over-allocating memory to an application. I thought this to be a very sensible suggestion, so I tried it and here is what happened.

I booted with VM off and OT loaded--ATM reported the LUB as 46 MB. I opened a PPP connection and launched Netscape 3.0b5 with every pluggin that I have and it claimed 12 MB. When I quit and closed the PPP connection, ATM reported 34 MB as the LUB rather than 46 MB, i.e. the 12 MB used by Netscape was "gone". I used Get Info to allocate 45 MB to MS Word 5.1 and launched it. ATM reported MS Word as claiming only 34 MB, not the 45 MB I had assigned it. The LUB was just 348K so the 12 MB was still "missing". I then used Get Info to allocate 12 MB to Wordperfect 3.5 and launched it with Word still open. I expected a crash but no crash occurred. ATM reported Wordperfect as claiming 12 MB with 248K left over as the LUB. The 12 MB was "back" in the allocation usages of 34 MB for Word and 12 MB for Wordperfect. However, after quitting Word and Wordperfect, ATM still reported the LUB as only 34 MB.

The results of this experiment are interesting and one might conclude that ATM is simply not correctly reporting the available memory. However, others have reported low memory error messages. Furthermore, the applications only claimed memory allocations and did not actually try to use them. With 56 MB of physical RAM that is not easily removed, I don't have enough RAM intensive graphics applications to easily test actual RAM use in the same manner. However, out of curiosity, I am going to find a way and will report the results in a future revision of this note.

I also experience this problem on an 8500/120 and a Quadra 605 using 7.5.3 and TCP/IP/OT. I have reverted to using MacTCP v2.0.6 on the Q605 which eliminates the RAM loss problem reported by ATM. I am going to try reverting to MacTCP on my 7100 to see if I can eliminate the PowerPC factor.

VI. Acknowledgement (8/29/96):

Rich Kubota of Apple's Open Transport development technical support team corresponded with me regularly regarding this problem over the past three weeks as they were revising OT 1.1.1b6 to OT 1.1.1b7. Whatever they did between beta versions appears to have mostly fixed this annoying and potentially harmful problem when OT 1.1.1b7 is used in combination with OT/PPP1.0f1c6 instead of FreePPP. Many thanks!! I look forward to the final release.

Regards,

Bill Fox
For MacOS Users Only!

[Back to top of page]





Copyright © 1997 by Bill Fox.
All rights reserved--no commercial use permitted without my expressed permission.

Of course . . . . . . and I am proud to be an Evangelista! . . .

[Back to top of page]