Re: [9fans] strangeness on a server

Re: [9fans] strangeness on a server

[erik quanstrom]

> Nov 29 17:57:56 EST 2009 local.c 5559 [bootes]
> local.c:227 c /n/dump/2009/1129/sys/src/9/boot/local.c:227
> < 			run("/boot/venti", "-c", f[0], "-m", "20", "-a", f[1], "-h", f[2], 0);
> ---
> > 			run("/boot/venti", "-c", f[0], "-a", f[1], "-h", f[2], 0);
>
> And things are *much* better.
>
> I wonder if this should not be the default for local.c.

that makes sense.  the problem i have is that venti is
either trying too hard to fix bad defaults, or not hard
enough.  i think it should just do what we tell it and
crash if it gets unreasonable input.

- erik

Re: [9fans] strangeness on a server

[erik quanstrom]

> one thing that I don't understand: when it starts up I get a line like this:
> mem -1 bcmem 0 icmem 0
>
> Does this point to misconfiguration or will venti do something reasonable here?

it looks like a little bit of misconfiguration, which the current
code makes too easy.  i'm not sure this is the root cause of
your problems.  since venti also does a big dance to try to make
memory misconfiguration non-fatal.

did you upgrade the hardware?  i have some tricks in
9atom (source ftp://ftp.quanstro.net/other/kernel.mkfs.bz2)
in particular, /dev/irqalloc has an extra field that will tell you
how many interrupts a particular vector has taken.  this
can be useful in narrowing down the guilty parties.

- erik

/n/dump/2009/1129/sys/src/cmd/venti/srv/venti.c:54,59 - /sys/src/cmd/venti/srv/venti.c:54,70
  }

  void
+ minmem(u32int *mem, u32int *icmem, u32int *bcmem)
+ {
+ 	if ((int)*icmem < 6 * 1024 * 1024)
+ 		*icmem = 6 * 1024 * 1024;
+ 	if ((int)*mem < 1 * 1024 * 1024)		/* lumps */
+ 		*mem = 1 * 1024 * 1024;
+ 	if ((int)*bcmem < 2 * 1024 * 1024)
+ 		*bcmem = 2 * 1024 * 1024;
+ }
+
+ void
  usage(void)
  {
  	fprint(2, "usage: venti [-Ldrsw] [-a ventiaddr] [-c config] "
/n/dump/2009/1129/sys/src/cmd/venti/srv/venti.c:184,195 - /sys/src/cmd/venti/srv/venti.c:195,201
  			mem = avail;
  			bcmem = 2 * avail;
  		}
- 		if (icmem < 6 * 1024 * 1024)
- 			icmem = 6 * 1024 * 1024;
- 		if (mem < 1 * 1024 * 1024)		/* lumps */
- 			mem = 1 * 1024 * 1024;
- 		if (bcmem < 2 * 1024 * 1024)
- 			bcmem = 2 * 1024 * 1024;
+ 		minmem(&mem, &icmem, &bcmem);
  		config.mem = mem;
  		config.bcmem = bcmem;
  		config.icmem = icmem;
/n/dump/2009/1129/sys/src/cmd/venti/srv/venti.c:201,207 - /sys/src/cmd/venti/srv/venti.c:207,214
  		bcmem = config.bcmem;
  	if(icmem == 0)
  		icmem = config.icmem;
- 	fprint(2, "%s: mem %d bcmem %d icmem %d...", argv0, mem, bcmem, icmem);
+ 	minmem(&mem, &icmem, &bcmem);
+ 	fprint(2, "%s: mem %ud bcmem %ud icmem %ud...", argv0, mem, bcmem, icmem);

  	if(haddr == nil)
  		haddr = config.haddr;

Re: [9fans] strangeness on a server

[ron minnich]

On a hint from Geoff I did this:

Nov 29 17:57:56 EST 2009 local.c 5559 [bootes]
local.c:227 c /n/dump/2009/1129/sys/src/9/boot/local.c:227
< 			run("/boot/venti", "-c", f[0], "-m", "20", "-a", f[1], "-h", f[2], 0);
---
> 			run("/boot/venti", "-c", f[0], "-a", f[1], "-h", f[2], 0);

And things are *much* better.

I wonder if this should not be the default for local.c.

ron

[9fans] strangeness on a server

[ron minnich]

OK, I'm seeing the following: 1500 syscalls/second, steady disk
activity, pegged really, while I'm not really doing anything, tends to
climb to 10000 (!). This just started 2 days ago when I upgraded ...
it had been a while. dma is enabled.

I'm suspecting venti or fossil or interaction but ... what's a good
way to tell. I also just realized that this is an old disk and I'm
running new venti. I don't have a bloom filter. Will that cause huge
trouble -- it does not seem to have.

I assume it is still safe to write a new config to /dev/sdC0/arenas as
in old venti? I want to set up the http port on venti to a more useful
value for me.

one thing that I don't understand: when it starts up I get a line like this:
mem -1 bcmem 0 icmem 0

Does this point to misconfiguration or will venti do something reasonable here?

thanks

ron