Re: MM managers and win98 compatibility

> I will bite the bullet and run all the MM's under Windows 98 today and let
> you guys know my results.

The following are my test results under Windows 98 SE. Keep in mind that my
only access to Windows 98 is running under VMWare with 256 meg of memory
reserved, so some of these tests *may* be influenced by that. However, this
should give a good starting point for the authors of the different memory
managers to investigate performance and failure under Windows 9x.

The following memory managers passed all validations:

BucketMM, EWCMM, FastMM, FastMM3 (3.01, found on b.p.attachments), PSDMM,
PureVCL, RecyclerMM, TopMM, WinMem

The following memory managers didn't pass all validations:

MultiMM
Validation:
Running Validate13...
Running Validate14... - Failed: Validate13;
Running Validate17... - Failed: Validate13;Validate14;
Failed Validations: Validate13;Validate14;

NexusMM
Validation:
Running Validate14...
Running Validate17... - Failed: Validate14;
Failed Validations: Validate14;Validate17;

QMemory
Validation:
Running Validate17...
Failed Validations: Validate17;

Now, regarding benchmarks run under Windows 98 SE, the following memory
managers were able to successfully complete all benchmarks (with the
exception of the 4 and 8 thread Nexus DB tests, which I disabled due to time
constraints):

BucketMM, EWCMM, FastMM, FastMM3 (3.01), PSDMM, PureVCL

The following memory managers had problems running the noted benchmarks
under Windows 98 SE. Many of the problems were that given benchmarks
"never" seemed to complete. The ones I note which didn't complete were
given at least five minutes, after which I stopped them due to time
constraints. As I stated, this is under VMWare, which *may* be the problem,
but the above memory managers completed just fine.

MultiMM:
Benchmarks which didn't complete: Block downsize, Small downsize, Block size
spread, Address space creep, Large block spread, Replay: eLink, Replay: COM+
Server.
The following benchmarks, if checked and run, consistently resulted in the
following error:
Raw performance 1 thread +
Array upsize 1 thread +
Many short lived threads =
FASTCODEMMCHALLENGEBV_MULTIMM caused an invalid page fault in
module FASTCODEMMCHALLENGEBV_MULTIMM.EXE at 0167:004061e8.
Registers:
EAX=004086f8 CS=0167 EIP=004061e8 EFLGS=00010207
EBX=318d0874 SS=016f ESP=318cf5e8 EBP=318d0774
ECX=00000000 DS=016f ESI=81960324 FS=675f
EDX=318d0770 ES=016f EDI=318d0874 GS=0000
Bytes at CS:EIP:
50 8b f2 8b d8 85 db 74 3d 81 7b 04 00 00 01 00
Stack dump:

Also of note, MultiMM doesn't seem to free memory properly under Windows 98.
This can be seen by running the Array upsize 1 thread benchmark and noting
memory usage afterwards. Also, note the time taken if only Nexus DB 1
thread is run, then run Array upsize 1 thread followed by Nexus DB 1 thread,
and notice the time increase.

NexusMM:
Benchmarks which didn't complete: Small upsize, Array upsize
Raw performance 1 thread resulted in an access violation
Large block spread resulted in an out of memory error

QMemory:
Benchmarks which didn't complete: Replay: eLink
Large block spread resulted in an out of memory error

RecyclerMM:
Benchmarks which didn't complete: Array upsize

TopMM:
Array upsize caused:
FASTCODEMMCHALLENGEBV_TOPMM caused an invalid page fault in
module KERNEL32.DLL at 0167:bff7a388.
Registers:
EAX=7c50f008 CS=0167 EIP=bff7a388 EFLGS=00010202
EBX=00000010 SS=016f ESP=00abf338 EBP=00abf35c
ECX=00000000 DS=016f ESI=7e770004 FS=2f57
EDX=014f0000 ES=016f EDI=0044bfec GS=0000
Bytes at CS:EIP:
89 41 08 8b 56 04 8b 46 08 89 50 04 8d 04 1e 50
Stack dump:
0044bfec 00000000 0225fffc 0226000c 7c510008 00000000 00000000 0007ebbb
0007e771 00abf380 bff7e338 014f0000 7e770004 00000010 00000002 02260008

Many short lived threads caused: Not enough storage is available to process
this command.

WinMem:
Benchmarks which didn't complete: ReallocMem, Small upsize, Small downsize

And those are the results of my tests under Windows 98 SE. However, I reran
these tests under Windows XP and found that some of the errors occurred
there as well, namely the AV during raw performance tests with NexusMM and
the out of memory error during the large block spread benchmark with
QMemory.


.

Relevant Pages

  • RE: Virtual memory
    ... You can configure virtual memory in windows xp and thereby improve the ... How to set performance options in Windows XP ... if you have background programs such as printing or disk ... It is also known as the paging file. ...
    (microsoft.public.windowsxp.perform_maintain)
  • [NT] NNTP Service in Windows Contains Memory Leak
    ... NNTP Service in Windows Contains Memory Leak ... An affected server could be restored to normal service by ...
    (Securiteam)
  • Re: Physical Memory
    ... These Windows services are started: ... Memory optimizers/defragers are nothing more than snake ... These optimizers work by making demands on the Windows Memory Manager ... The funny thing about all of this nonsense is that the snake oil memory ...
    (microsoft.public.windowsxp.general)
  • Re: Slow performance
    ... computer has two memory slots so to achieve 1 gb you need to install 2 x ... own set up consists of AVG 8 Anti-Virus, Spybot S& D, Windows Defender ... of programme is generally not recommended in these newsgroups. ... My disk drive is pretty full right now, mainly due to fact that I just ...
    (microsoft.public.windowsxp.perform_maintain)
  • Re: Physical Memory
    ... Thanks for your help, John. ... As for the services just use the Windows Services Management Console to ... Remote Access Connection Manager ... gig Pentium 4 processor and apparently a half gig of available memory, ...
    (microsoft.public.windowsxp.general)
Loading