Memory / Handle Leak in TCserver and TabTip
Dino
I've recently found out that TCserver & TabTip have been hoarding memory on
my machine; I also found out that it is not a new issue.
What I do have to add to this topic is what I've noticed when checking the
status of these two processes with Task Manager, and Process Explorer. After
having my laptop running non-stop for many days, the memory usage of TCserver
went over 320MB; in Task Manager I noticed that it had more than 400 thousand
handles open!! Using process explorer I found out that the bulk of those
handles are of the Thread type, while the actual running threads - according
to Task Manager - was around 100.
In this newsgroup I've stumbled upon another thread that had a link to an
interesting article
(http://blogs.msdn.com/jonathanh/archive/2004/08/29/222543.aspx), which
pointed out that these two are dot NET applications; having experience with
another "managed" language (Java), this seems to me like someone is still
keepin referencing .NET Thread objects, which in turn probably contain Win32
Thread handles. Having noticed that the number of active threads in TCserver
fluctuates, as applications are run & terminated, then it must be spawning
new threads as needed; I can definetly see a scenario in which Thread objects
are placed in a link list, and then forgotten.
Additionally, the primary applications I've been using during that time were
Firefox, Thunderbird, and mplayer2; no inking at all.
I didn't check that status of TabTip as thoroughly as TCserver, but I do
remember that it was using a healthy chunk of memory. For next time check
both, as well the .NET counters mentioned in the article above.
Finally, since I marked this as a question - is Microsoft doing anything
about this? In the other two posts I stumbled upon, a MVP mentioned that this
will not be fixed; is this true? This is definetly a leak, which leads to an
unuasble machine; and asking the user to restart those two processes seems
like unacceptable. Can someone with contacts at Microsoft get this info about
the Handles to them?
Thanks.
Chris H.
currently, as you noted from the blog information from August, 2004. As I
understand it, until there's a complete revision (Longhorn timeframe,
possibly) this is the way it is.
I think it is important to understand, though, as pointed out in the blog,
that the amount of "memory" you're seeing is listed the way the NT handles
memory-mapped files - they are "committed memory" not actual memory or
pagefile space in use.
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Dino" <Di...@discussions.microsoft.com> wrote in message
news:2881B0A3-4C6E-467A...@microsoft.com...
Mike Williams [MVP]
> Finally, since I marked this as a question - is Microsoft doing anything
> about this? In the other two posts I stumbled upon, a MVP mentioned that this
> will not be fixed; is this true? This is definetly a leak, which leads to an
> unuasble machine; and asking the user to restart those two processes seems
> like unacceptable. Can someone with contacts at Microsoft get this info about
> the Handles to them?
I've seen the "explanation" from someone at Microsoft as to why it
doesn't need to be fixed, but frankly it doesn't hold water for me. When
these processes get over 100MB of RAM, my machine becomes almost unusable.
As a result, I run this batch file once per week (thanks to the person
who posted this before - it's a godsend). Machine performance is
instantly restored.
****************************
REM - First, kill processes
taskkill -im tabtip.exe /F
taskkill -im tcserver.exe /F
REM - switch to directory containing Ink executables
c:
cd "C:\Program Files\Common Files\Microsoft Shared\Ink\"
REM - Startup the processes -
REM - tcserver.exe will be called automatically by tabtip.exe
tabtip.exe
******************************
Mike
Dino
operating system threads. According to the .Net info from Process Explorer:
# of current logical Threads: 1,239
# of current physical Threads: 907
# of current recognized threads: 336
# of total recognized threads: 1,078
Furthermore, the GC information says:
# Gen 0 Collections: 134
# Gen 1 Collections: 34
# Gen 2 Collections: 8
I wrote a small program program to understand how these number play out, and
the thing that baffles me is the fact that the # of physical threads does not
correspond to the number of operating system threads (16) with the
tcserver.exe.
In my test program, I create C# Thread objects, that do nothing but sleep
forever; the stats on it show a 1-to-1 correspondence between the physical
threads and operating system threads.
Now about the handles - in my program, when I modify the code so that the
created C# threads to finish immediately, then the operating system's threads
handles are released, whether or not the C# Thread object is garbage
collected; this is obviously not the case with tcserver, meaning that the
logical (C#) threads are still active, and maintain all their resources. I
eyeballed the list of handles in Process Explorer, and I can say with some
degree of confidence that the number of logical threads (1,239) corresponds
to the number of Thread handles; there are plenty of other handles there,
probably used in some way by the logical threads.
Finally, a comment about garbage collection - as the numbers from Process
Explorer demonstrate, tcserver.exe went through 8 collections in Generation
#2; that means it had alraedy gone through 8 full grabage collections, so if
any of those threads were truly garbage, then they would've been collected
already. The notion that managed code will increase in size as long as there
is space strikes me as wrong; Sun's Java VM the default initial & maximum
heap sizes are conservative, and additional memory is allocated only after
full garbage collections do not reduce the amount of used memory under a
certain threshold. Naturally, all these parameters are adjustable.
What I'm trying to get at, is that the .NET VM cannot be radically
different, and is evident by the small memory footprint of the TCserver upon
restart. A process that starts as 10M and ends up at over 300M, while
apparently doing nothing - that is definetly a memory leak.
I don't about you, but I think this is a problem that needs to be addressed
before longhorn. It would be great if you could forward my "findings" to
Microsoft (seeing as you are an MVP, you can get a hold of them quicker than
I can). I'm sure that if they had a developer on this, he/she would have
realized this by now.
About the committed memory & handles - as far as I know, committed memory,
is memory that has space reserved for it in the swap file, i.e. actual memory
allocated; this is to be contrasted with reserved memory, which is only
virtual.
I might be a bit off on the definitions, but fact is that when Task Manager
indicates that TCserver has a 300M+ memory usage, then my laptop is quite
slow.
Chris H.
Pagefile. I'm not enough of a techie to explain it, but that is as
described in the blog URL I posted. I will try passing word along, though.
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Dino" <Di...@discussions.microsoft.com> wrote in message
news:6CEB6064-1632-43D2...@microsoft.com...
Dino
But, I think there is a difference between what the .NET Runtime considers
as reserved, and what the operating systems considers as reserved.
When I try to figure out how much memory a process uses, I don't really care
what the .NET numbers are, since they will be included in what the OS has to
say about the process as a whole.
So, at this very moment my TCserver has a Memory Usage of 37M, and a Virtual
Memory Size of 33M; the latter must be wrong, but more on that in a moment.
According to Task Manager's Help, the Memory Usage is "... the current
working set of a process, in kilobytes. The current working set of a process
is the number of pages currently resident in memory."
Process Explorer give the same value for Memory Usage / Working Set, while
the VM size is at 127M. This value for the VM size makes sense, so I wouldn't
trust Task Manager's VM values.
So, both programs say that the process is using 37M of physical/actual/real
memory, and I believe it. When TCserver Mem Usage was 300M+ on my 1GB tablet,
believe me, it was definetly used/committed.
So, to reiterate - the actual memory used is displayed by Task Manager in
the Memory Usage column, and the Working Set column in Process Explorer. In
order to check the Virutal Memory size, use Process Explorer.
Perhaps you can pass along the word that Task Manager is displaying the
wrong number for the VM Size?
Thanks.
Chris H.
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Dino" <Di...@discussions.microsoft.com> wrote in message
news:E4D0EBF5-4689-4221...@microsoft.com...
Dino
Thanks.
Micah Brodsky
it's simply that too few users experience this problem. I
just discovered it myself. Seven thousand kernel threads
in TCServer? Yeah. No wonder my mouse pointer was jerking
around like a drunkard.
--Micah
>.
>
Chris H.
Edition 2005? If so, have you upgraded your drivers, especially graphics
and pen, since installing the above? How much RAM do you have?
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Micah Brodsky" <anon...@discussions.microsoft.com> wrote in message
news:2a8401c50425$df83e030$a501...@phx.gbl...
Jason Matte
If you would open your mind and listen you may just be surprised at the
results.
Get a Mac. It works. It really does.
It doesn't crash. It doesn't have memory leaks. It is NOT prone to viruses,
spyware, malware, trojans, keyloggers, oh...did I mention viruses?
My uptime on my Powerbook 17 inch is currently sitting at 4 months!!!! I
have not powered it off once in 4 months!!!!!!!!!
Try that in Windows.
Good luck.
Follow my advice and you can stop wasting time posting crap like this that
NO ONE CARES about.
Jason
>.
>
Dino
Acer C300, with 1GB RAM; I applied SP2, hence I have the 2005 Edition. My
drivers are up to date, although I fail to see how it could be related to the
leak.
I'm guessing that other people didn't notice it because they probably
shutdown the machine most of the time; either that, or that they are the
"common user" which doesn't have a clue on what's going on, and simply
reboots, and everything is back to normal....
Woody
> I also have the problem discribed a few times a day.
>
> My tablet (HP TC1100) hangs and I can't do anything with it anymore.
> >
> I however hope this problem will be fixed soon...
So does anyone know the reasons that some do it and some don;t? I
haven't seen this on my TC1100, and I use it extensively
--
Woody
Woody
> I have the simplest solution for all of you.
> If you would open your mind and listen you may just be surprised at the
> results.
> Get a Mac. It works. It really does.
I have 4 macs. How does that help a memory leak in my TIP?
> Follow my advice and you can stop wasting time posting crap like this that
> NO ONE CARES about.
I think you will find that in a tablet PC group, that is probably what
you appear to be doing.
I love my Macs, I post about them in the Mac groups. It is more
constructive that way.
--
Woody
Woody
> I wonder if I mentioned in my other messages in this thread, so here it is:
> Acer C300, with 1GB RAM; I applied SP2, hence I have the 2005 Edition. My
> drivers are up to date, although I fail to see how it could be related to the
> leak.
A driver has a lot to do with the potential for leaks I would have
thought.
> I'm guessing that other people didn't notice it because they probably
> shutdown the machine most of the time; either that, or that they are the
> "common user" which doesn't have a clue on what's going on, and simply
> reboots, and everything is back to normal....
Speaking as one of these 'common users', a Software engineer building
distributed network documentation systems, My tablet was last rebooted
on tuesday, and has been used quite a lot in the 5 days since then. My
tabtp is at 24M (admitadly too high) with 19 threads.
I have never seen this problem that a small number of people are
describing so I was wondering what combination of factors causes it.
--
Woody
www.alienrat.com
Tried Yagol for tablet yet? : http://www.alienrat.com/games
Mike Williams [MVP]
AFAICT from official responses, the Tablet team have decided that it's a
non-issue and are no longer investigating :-(.
Woody
says...
That is clearly wrong then. It appears it is a problem for some people,
and with it splashed over the register, I would have thought it was an
issue for the tabletPC team in general.
--
Woody
Played Yagol for tabletPC yet?
http://www.alienrat.com/games
Woody
Chris H.
Tablet Team at Microsoft is looking into it.
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Woody" <use...@alienrat.co.uk> wrote in message
news:MPG.1c6806ba5...@192.168.254.11...
Dino
> A driver has a lot to do with the potential for leaks I would have
> thought.
That's a pretty broad statement. I think Chris asked about the drivers in
order to cover the essentials of troubleshooting; and I'm usually on the
bleeding edge when it comes to drivers.
>
> > I'm guessing that other people didn't notice it because they probably
> > shutdown the machine most of the time; either that, or that they are the
> > "common user" which doesn't have a clue on what's going on, and simply
> > reboots, and everything is back to normal....
>
> Speaking as one of these 'common users', a Software engineer building
> distributed network documentation systems, My tablet was last rebooted
> on tuesday, and has been used quite a lot in the 5 days since then. My
> tabtp is at 24M (admitadly too high) with 19 threads.
>
> I have never seen this problem that a small number of people are
> describing so I was wondering what combination of factors causes it.
Heh, hope you didn't get offended :D
Although you probably don't fit my definition of a "common user".
What was your tcserver.exe status? on my machine it dwarfed tabtip.exe.
From what I can tell, certain applications cause the leak to exacerbate. I
think Firefox has a negative effect on tcserver.exe, since I've seen it
misbehaving in regards to ink input; for example - while hovering with the
pen over the address bar, I would get that popup button that allows me to
open the TabletPC input panel; however, this would only happen once, and
never again.
I've observed this behaviour long before I noticed the leak, and I knew that
there is a problem when it comes to Firefox and the "Advanced Text/Input
Services"; based on that knowledge I disabled these services (in the Firefox
shortcut), and apparently tcserver.exe's leakage seems to have slowed down.
Still, I wouldn't swear that this is a good workaround, until I tested it
some more.
In any case, they should still fix the the problem at the source.
What kind of applications are you using?
Chris H.
works in all applications (such as Firefox). It doesn't because not all
apps are written with context tagging, i.e., identifying themselves as a
place where input is accepted. Firefox, along with several others, is not
"current" for Ink awareness unless you use a third-party app which adds the
tagging.
You'll note the Office XP (2002 versions) which came out before the Tablet
PCs hit the market, are functional only with the Office XP Pack for Tablet
PC, and even though Office 2003 has much improved (native tagging) there are
additional areas where Microsoft has added in items:
http://www.microsoft.com/downloads/details.aspx?FamilyID=31634d79-34ad-494d-8108-80085ace23be&displaylang=en
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Dino" <Di...@discussions.microsoft.com> wrote in message
news:8EA70B08-D5F5-4F7D...@microsoft.com...
Woody
Di...@discussions.microsoft.com says...
> "Woody" wrote:
> > > I'm guessing that other people didn't notice it because they probably
> > > shutdown the machine most of the time; either that, or that they are the
> > > "common user" which doesn't have a clue on what's going on, and simply
> > > reboots, and everything is back to normal....
> >
> > Speaking as one of these 'common users', a Software engineer building
> > distributed network documentation systems, My tablet was last rebooted
> > on tuesday, and has been used quite a lot in the 5 days since then. My
> > tabtp is at 24M (admitadly too high) with 19 threads.
> >
> > I have never seen this problem that a small number of people are
> > describing so I was wondering what combination of factors causes it.
>
> Heh, hope you didn't get offended :D
Been on usenet too long to get offended by anything!
> Although you probably don't fit my definition of a "common user".
I guess not!
> What was your tcserver.exe status? on my machine it dwarfed tabtip.exe.
tcserver was quite a way below tabtip.exe. At the time I looked I sorted
by size and tabtip was the largest process. tcserver was quite a way up
the list.
> From what I can tell, certain applications cause the leak to exacerbate. I
> think Firefox has a negative effect on tcserver.exe, since I've seen it
> misbehaving in regards to ink input; for example - while hovering with the
> pen over the address bar, I would get that popup button that allows me to
> open the TabletPC input panel; however, this would only happen once, and
> never again.
I get the popup button which lets me open the panel reliably, but the
panel goes away after one or two keypresses on the keyboard (it stays on
the ink part though).
> I've observed this behaviour long before I noticed the leak, and I knew that
> there is a problem when it comes to Firefox and the "Advanced Text/Input
> Services"; based on that knowledge I disabled these services (in the Firefox
> shortcut), and apparently tcserver.exe's leakage seems to have slowed down.
> Still, I wouldn't swear that this is a good workaround, until I tested it
> some more.
> In any case, they should still fix the the problem at the source.
>
> What kind of applications are you using?
At the time of writing I had firefox, acrobat reader, Agilix GoBinder,
Paint.net, eudora, Yagol, MS Visual Studio.net and Excel open.
Firefox, goBinder and eudora are mostly permanently running, then there
is office, VS.net and others that come and go as required.
Dino
doesn't interact nicely with the TIP, and apparently contributes to the
leakage rate. If I were to debug the leakage problem, then I would probably
use Firefox in the process. That's all.
Good to learn some TPC lingo, though.
And thanks for the link.
Chris H.
threads previously in this newsgroup) with Firefox, that application itself
is not Ink aware. It wasn't written for use on a Tablet PC by the software
developers, so digging into the Tablet Input Panel for the "error" is
looking in the wrong direction. It is Firefox itself, just like problems
people are reporting with Adobe. Until they take into account people are
using their programs on Tablet PC with pens, the failure is with the app
itself.
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Dino" <Di...@discussions.microsoft.com> wrote in message
news:A6103FBC-3FCB-4975...@microsoft.com...
Woody
says...
> Dino, the point is, unless you're using the add-on (mentioned in several
> threads previously in this newsgroup) with Firefox, that application itself
> is not Ink aware. It wasn't written for use on a Tablet PC by the software
> developers, so digging into the Tablet Input Panel for the "error" is
> looking in the wrong direction. It is Firefox itself, just like problems
> people are reporting with Adobe. Until they take into account people are
> using their programs on Tablet PC with pens, the failure is with the app
> itself.
>
So you are saying that it is firefoxs fault that another program starts
bloating and using memory/threads incorectly, even though firefox knows
nothing about it?
however, I think this is a red herring as I use firefox exclusively and
don't get the problem, unless it is an interaction between firefox, the
tip and something else.
Dino
working properly with the "unaware" Firefox; I'm only remarking about it's
apparent effect on tcserver.exe.
So, by "debugging" I meant to investigate the effects of the unware Firefox
on tcserver.exe, and not the bizzare TIP behaviour.
Chris H.
using the third-party add-on which enables context tagging in the input
line, whatever you're trying to do is going to fail or cause other issues
(perhaps like what you're seeing).
If I use my snow thrower to clear the driveway after a fresh snowfall, it
does the job as designed. However, if I attempt to clear snow left over
night and rained upon, the snow thrower will fail, although it continues to
try. 8-)
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Woody" <use...@alienrat.co.uk> wrote in message
news:MPG.1c68778e9...@192.168.254.11...
Woody
winx...@hotmail.com says...
> What I'm saying is Firefox was not designed as "Ink aware" and unless you're
> using the third-party add-on which enables context tagging in the input
> line, whatever you're trying to do is going to fail or cause other issues
> (perhaps like what you're seeing).
I know what you are saying, but if this were the case, it would indicate
a further problem with the tip, in that it couldn't cope with non ink
aware programs.
>
> If I use my snow thrower to clear the driveway after a fresh snowfall, it
> does the job as designed. However, if I attempt to clear snow left over
> night and rained upon, the snow thrower will fail, although it continues to
> try. 8-)
But will your snow thrower bloat up to many times its size and then fail
in that situation?
Chris H.
compressed ice, necessitating the application of a very large flathead
screwdriver to clear it. 8-)
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Woody" <use...@alienrat.co.uk> wrote in message
news:MPG.1c687f1ee...@192.168.254.11...
Woody
> The snow thrower "chute" does clog up, yes. 8-) Packs that sucker full of
> compressed ice, necessitating the application of a very large flathead
> screwdriver to clear it. 8-)
Ahh - shows how much I know about snow blowers. When it snows here it
gets on the news (as everything stops!)
--
Woody
Chris H.
--
Chris H.
Microsoft Windows MVP/Tablet PC
Tablet Creations - http://nicecreations.us/
Associate Expert
Expert Zone - www.microsoft.com/windowsxp/expertzone
"Woody" <use...@alienrat.co.uk> wrote in message
news:1gr9m5q.1biias1n7bitbN%use...@alienrat.co.uk...
Woody
> That's what you get for not living where there are four seasons. 8-)
We have them:
Cold and wet,
cool and wet,
hot and dry with water shortages,
warm, windy and wet with floods.
What more do you need?
--
Woody
Dino
trigger in tcserver.exe's leak.
Here are the results:
Upon rebooting the machine, tcserver.exe's memory usage was about 14M, with
617 handles. Then I popped a Firefox window, went to a website, and started
minimizing & restoring it repeatedly; the number of handles grew to 2,000 to
3,000, but after a certain point, it would drop to around 500 - probably due
to a garbage collection.
After my finger got tired from the repetitive clicking, I wrote two HTML
pages with some JavaScript (I'll paste the code below), in order to pop some
windows, and then close them. After a certain point, tcserver.exe started
bloating again, and the handle count went up to 25,000. At this point I
terminated Firefox, modified the shortcut to disable the advanced text
services, and run the test again; this time the handle count didn't budge,
while previously it would increase by ~25 handles per popped window.
However, as I noted above, Firefox is really not to blame for anything; I
believe I proved by the next step: do the same test with Internet Explorer
(without reboot the machine) - the leakge continued.
My guess/conclusion is that at some point tcserver.exe looses track of it's
resource, and then it starts leaking. After that point, there is no way for
it to recover.
The HTML pages are below; save them in a directory, and then open index.html.
This is the only test I could think of that wouldn't take too much time and
effort.
---------------------------- index.html
---------------------------------------
<html>
<head>
<script language="JavaScript">
var win1;
var win2;
var win3;
var win4;
function openEm ()
{
win1 = window.open("pop.html");
win2 = window.open("pop.html");
win3 = window.open("pop.html");
win4 = window.open("pop.html");
setTimeout('closeEm()', 2000);
}
function closeEm ()
{
win1.close();
win2.close();
win3.close();
win4.close();
setTimeout('openEm()', 100);
}
</script>
</head>
<body onLoad="window.setTimeout('openEm()', 100);">
<p>hello</p>
</body>
</html>
--------------------------- pop.html -----------------------------------
<html>
<head>
</head>
<body>
yelllow
<input>
<input>
<input>
</body>
</html>
Mike Williams [MVP]
> I did some testing, and perhaps I was too quick to point to Firefox as the
> trigger in tcserver.exe's leak.
Firefox has its own memory problems independent of those in
Tabtip/TCServer(which I was experiencing long before I installed Firefox).
The tablet process leaks seem to be exascerbated by using the pen in MSN
Messenger and Office Document Imaging programs. I almost never use the
tip, so it's interesting to see its usage climb steadily in a session
where it has never been invoked.
Woody
Well, I never use MSN messenger, and although I use office, I am not
sure I use Office Document imaging, unless it is used in the normal
course of using office.
--
Woody
CraigB1009
tired of killing these memory hog processes
Reed [MVP]
> Im
> tired of killing these memory hog processes
Word from Microsoft as of today is that a fix has been developed and is
currently in testing. This in response to a similar question in today's
monthly Tablet PC Ask the Experts chat.
-Reed
MVP