[vwnc] [Bug] UHE saving large image to network drive

> -----Ursprüngliche Nachricht-----
> Von: [hidden email]
> [mailto:[hidden email]] Im Auftrag von Valloud, Andres
> Gesendet: Dienstag, 19. August 2008 09:01
> An: [hidden email]
> Betreff: Re: [vwnc] [Bug] UHE saving large image to network drive
>
> Christian,
>
> We have just characterized the issue more precisely.  Writes
> over the network may fail when the write block is too large
> because then Windows fails to allocate enough memory to
> buffer the write.  Using unbuffered writes is not an option
> we would like to consider because, besides writes being
> slower when done locally, it has several additional
> restrictions which would complicate the implementation of the
> VM.  Instead, the interim solution we just verified to work
> is to recursively halve the write block until writes succeed,
> and to do this only when receiving error
> ERROR_NO_SYSTEM_RESOURCES (1450).
>
> Note that the fix described in the paragraph above, which we
> just tested and verified to work, has not yet gone through
> our verification process.
>
> This also shows that the image size is not the trigger of
> this problem.  Rather, what causes the issue to crop up is
> the size of memory segments allocated by the VM.  So, in
> other words, a 400mb image with 10mb old space segments would
> most likely save successfully, while a 200mb image with a
> 100mb object in it runs a higher risk of failing to save over
> a Windows network to a shared drive.
>
> Andres.
>
>
> -----Original Message-----
> From: [hidden email]
> [mailto:[hidden email]] On Behalf Of Christian Haider
> Sent: Saturday, August 09, 2008 2:51 AM
> To: [hidden email]
> Subject: Re: [vwnc] [Bug] UHE saving large image to network drive
>
> I did some more experiments (without success though):
>
> - both client and server have an 1GB card, but the routers
> can only handle 100MB. So I connected the client directly
> with the server giving me a 1GB LAN connection (but I'm sure
> the cable cannot handle the full speed). - same error
> - used UNC filename instead of the mapped drive - same error
> - looked at the file handles as Eliot suggested (I'm not sure
> how though). TaskManager shows 86 Handles (I guess that are
> files and sockets). After doing 'IOAccessor closeAll' the
> number dropped from 86 to 79. - same error
> - closing all windows and doing several GCs - same error
> - saving to a 8.3 compatible filename (s:\image\a.im) - same error
> I would rule out a filename problem since parts of the
> file (around 118 MB of the 149 MB) were written before the
> error occured.
> - saved to a network path where the disk does not have a
> shadow copy - same error
> - closing all programs (Outlook and Firefox) (couldnt stopp
> AVG virus scanner though) - same error
> - saving to another client computer instead to the server - same error
> - saving from another WinXP client - same error
> - starting and saving locally on the server (win2003) - works
> - starting on the server and saving across the LAN - same error
> - Assert and debug VMs dont show any problems.
>
> Here is a script for a reproducible testcase (important is
> the network path!:
>
> ObjectMemory globalGarbageCollect.
> ((1 to: 1650) collect: [:a | (1 to: 10000) collect: [:b |
> b]]) inspect.
> ObjectMemory snapshotAs: 's:\image\a1650' thenQuit: false
>
> DoIt in the workspace of the vanilla visual.im. For me the
> error occurs with 1650 and above. 1649 and below works fine.
> The saved image size with 1649 is 77880 KB.
> It would be interesting if others can reproduce the error
> (probably with a different number).
>
> Preliminary conclusions:
> - it has to do with the LAN
> - it has to do with the image size (not content, I think)
> - it might be a problem of windows (XP and 2003)
> - the problem seem to be network writes of the VM
>
> Hth,
> Christian
>
> > -----Ursprüngliche Nachricht-----
> > Von: [hidden email]
> > [mailto:[hidden email]] Im Auftrag von Steven Kelly
> > Gesendet: Freitag, 8. August 2008 11:01
> > An: [hidden email]
> > Betreff: Re: [vwnc] [Bug] UHE saving large image to network drive
> >
> > Paul Baumann wrote:
> > >
> >
> '\\rv-bp-iffs-01\migrate_server\Smalltalk\builds\12.200\12.200 053.0\'
> > > asFilename withSeparator.
> > >
> > > a FATFilename('\\rv-bp-iffs-
> > > 01\migrate_server\smalltal\builds\12.200\12200053.0\')
> >
> > I've seen this problem occasionally over the years, when
> working with
> > any network file (maybe only UNCs, not mapped drives). VW doesn't
> > recognize the UNC network disk as NTFS and reverts to
> FATFilename and
> > truncates the directory and file parts to 8.3. Most problems
> > disappeared when FATFilename was corrected (5i.*?) to
> default to long
> > (254 character) filename components when it couldn't get
> the correct
> > result from the OS call in getFileSystemAttributes:. Still,
> > occasionally we see the error: maybe getFileSystemAttributes:
> > in the VM still has some old default code? Line 1081 in 7.6
> ntfile.c
> > seems to set the file type to FAT and maximum file
> component length to
> > 12 (8.3) if GetVolumeInformation fails but the drive is not
> invalid,
> > and then returns the result to the image as a valid set of
> file system
> > attributes.
> > Presumably that's what's happening here.
> >
> > HTH,
> > Steve
> >
> > > -----Original Message-----
> > > From: [hidden email]
> [mailto:[hidden email]] On
> > > Behalf Of Valloud, Andres
> > > Sent: Thursday, August 07, 2008 2:45 AM
> > > To: [hidden email]
> > > Subject: Re: [vwnc] [Bug] UHE saving large image to network drive
> > >
> > > Christian,
> > >
> > > Initial examination reveals that indeed, as Paul Baumann
> commented,
> > > this is an issue of Windows itself running out of
> > resources.  I am not
> > > sure how the VM should be instructed to cope with the OS
> telling it
> > > there is no way to perform what it has been requested.  To
> > some extend
> > > I'd be inclined to think that a retrying strategy could work, but
> > > unfortunately error 1450 does not seem to necessarily
> > indicate that the
> > > error is transient.  In other words, at least in principle
> > there could
> > > be a valid reason for a 1450 error, and retrying might not
> > necessarily
> > > fix it nor be a reasonable course of action.
> > >
> > > >From the material I reviewed, at first sight I think that what is
> > > going on in your case is that the client box floods the 100mbps
> > > ethernet connection faster than it can write data to the server,
> > > eventually the client cannot allocate more memory for
> buffering the
> > > write, and then a 1450 error occurs.
> > >
> > > If you save an image with the task manager open, would you mind
> > > checking what are the values related to the Physical
> Memory and the
> > > Kernel Memory at the point when the VM reports a failure?
> > You may want
> > > to take a screenshot of the task manager at the point when
> > the debugger
> > > comes up so it's easier to take a look at things.  Feel
> > free to send it
> > > directly to me.
> > >
> > > Also, are you booting the client Windows with the /3GB
> switch?  This
> > > has the effect of limiting the address space of the OS to
> > 1GB, and in
> > > turn can cause errors like 1450 to occur more frequently.
> > >
> > > Apparently, a way to avoid this is to request that
> Windows does not
> > > buffer writes to the image file as they happen.  However,
> this will
> > > almost certainly have the side effect of making image writes
> > > considerably slower.  For example, without buffering,
> writes from a
> > > client over to a server will block the image saving process
> > until the
> > > server acknowledges that the write was committed to disk.  
> > This implies
> > > a network round trip per file write operation.  Even
> locally, this
> > > would potentially make writing image files take
> > considerably more time.
> > >
> > > Note that we have not yet decided exactly what to do with
> > this, and we
> > > are not done with our research yet either.
> > >
> > > Andres.
> > >
> > >
> > > -----Original Message-----
> > > From: Christian Haider
> > [mailto:[hidden email]]
> > > Sent: Wednesday, August 06, 2008 10:46 PM
> > > To: Valloud, Andres; [hidden email]
> > > Subject: AW: [vwnc] [Bug] UHE saving large image to network drive
> > >
> > > Thanks for looking into it.
> > >
> > > The system error looks like this:
> > > self:  a SystemError(#'io error',1450)
> > > parameter:  1450
> > > name:  #'io error'
> > >
> > > HTH,
> > >         Christian
> > >
> > > > -----Ursprüngliche Nachricht-----
> > > > Von: [hidden email]
> > > > [mailto:[hidden email]] Im Auftrag von Valloud, Andres
> > > > Gesendet: Donnerstag, 7. August 2008 05:30
> > > > An: [hidden email]
> > > > Betreff: Re: [vwnc] [Bug] UHE saving large image to
> network drive
> > > >
> > > > Christian,
> > > >
> > > > In order to debug this issue, we need to know what is
> the numeric
> > > > error that the VM reports.  From the stack you posted, the
> > > interesting
> > > > section to look at is this:
> > > >
> > > > SystemError>>handleErrorFor:
> > > > NTFSFilename(Filename)>>snapshot:
> > > >
> > > > In particular, the systemError object will have the
> > numeric error in
> > > > the instance variable called parameter.  Would you mind
> letting us
> > > > know what is the value of the systemError's parameter?
> > > >
> > > > Andres.
> > > >
> > > > ________________________________
> > > >
> > > > From: [hidden email]
> > > > [mailto:[hidden email]] On Behalf Of Christian Haider
> > > > Sent: Saturday, August 02, 2008 12:57 AM
> > > > To: vwnc List
> > > > Subject: [vwnc] [Bug] UHE saving large image to network drive
> > > >
> > > >
> > > > Usually, I work with my images on a network drive.
> > > > Unfortunately, when the image gets too large (only 105 or
> > 141 MB), I
> > > > get a UHE when trying to save the image (see attached).
> > > >
> > > > The exception reliably occurs in VW7.5 and 7.6.
> > > > Client WinXP sp3 connected with 100MB LAN to directory on
> > Win Server
> > > > 2003 with NTFS (no compression).
> > > >
> > > > "save image as..." and "save image" to a local drive
> > works flawlessly.
> > > >
> > > > The reason seems to be an IO error which is not
> adequately handled
> > > (by
> > > > the VM?).
> > > >
> > > > cheers,
> > > >     Christian
> > > >
> > > >
> > > > No virus found in this incoming message.
> > > > Checked by AVG - http://www.avg.com
> > > > Version: 8.0.138 / Virus Database: 270.5.12/1595 - Release
> > > > Date: 06.08.2008 08:23
> > > >
> > > No virus found in this outgoing message.
> > > Checked by AVG - http://www.avg.com
> > > Version: 8.0.138 / Virus Database: 270.5.12/1595 - Release Date:
> > > 06.08.2008 08:23
> > >
> > > _______________________________________________
> > > vwnc mailing list
> > > [hidden email]
> > > http://lists.cs.uiuc.edu/mailman/listinfo/vwnc
> > >
> > >
> > > This message may contain confidential information and is
> > intended for
> > > specific recipients unless explicitly noted otherwise. If
> you have
> > > reason to believe you are not an intended recipient of
> this message,
> > > please delete it and notify the sender. This message may
> > not represent
> > > the opinion of IntercontinentalExchange, Inc. (ICE), its
> > subsidiaries
> > > or affiliates, and does not constitute a contract or guarantee.
> > > Unencrypted electronic mail is not secure and the
> recipient of this
> > > message is expected to provide safeguards from viruses and pursue
> > > alternate means of communication where privacy or a binding
> > message is
> > > desired.
> > >
> > >
> > > _______________________________________________
> > > vwnc mailing list
> > > [hidden email]
> > > http://lists.cs.uiuc.edu/mailman/listinfo/vwnc
> >
> > _______________________________________________
> > vwnc mailing list
> > [hidden email]
> > http://lists.cs.uiuc.edu/mailman/listinfo/vwnc
> >
> >
> >
> > No virus found in this incoming message.
> > Checked by AVG - http://www.avg.com
> > Version: 8.0.138 / Virus Database: 270.5.12/1599 - Release
> > Date: 07.08.2008 20:49
> >
> No virus found in this outgoing message.
> Checked by AVG - http://www.avg.com
> Version: 8.0.138 / Virus Database: 270.5.12/1599 - Release
> Date: 07.08.2008 20:49
>
> _______________________________________________
> vwnc mailing list
> [hidden email]
> http://lists.cs.uiuc.edu/mailman/listinfo/vwnc
>
> _______________________________________________
> vwnc mailing list
> [hidden email]
> http://lists.cs.uiuc.edu/mailman/listinfo/vwnc
>
>
>