Mailing List archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[vdr] Re: 18 Minutes - what does that tell?



> Read the subject, and see what may happen if you don't watch that
> closely.

I don't know what's wrong with the '18-minute-system'. I just know, that I
used noflushd for months. I monitored the number of spin-ups to be
sure the HDD will not suffer. I never had more than appx. 10 spin-ups
a day. Usually, if there's no recording, it was sleeping for 600-1200 mins
until some read access forces it to spin up.

> > It's somewhere around 10, sometimes 20 cycles a day 
> > (depends on what else is running on your system). 
> > So 10-20 cycles a day makes appx. 7-13 years.
> 
> If someone does a lot of recordings and playbacks then 10-20 times is
> not much. But it also depends on the stop timeout you have programmed
> for the disk. If it is only 1 minute I am pretty sure that even *normal*
> use will cause much more spinups. If you have a timeout of 30 minutes it
> will be much less. This has certainly to be watched and tuned.

Correct. 1 minute timeout is nonsense. I found 15 minutes to be O.K. for
128MB RAM. With 512MB RAM the disk would eventually SPIN DOWN WHILE
RECORDING,
since 20-25 minutes of video fits into RAM. This results in the
buffer overflow errors while spinning up.
In theory this should not happen, noflushd is supposed to
write to disk as long as it is spinning (using a normal sized chache)
and start writing to the 'big' chache when the disk is not spinning.
Finally, spinning up the disk, when the big cache fills up.
Obviously something goes wrong with calculating (I don't know if this
is even implemented) the time when the cache is expected to be full in order
to spin up the HDD in time.
Since this behavior is unacceptable to me I don't use noflushd ATM.


> The other point I would like to mention is the data loss you will
> experience if the system crashes for some reason or there is a power
outage.

That will happen to a standard system, too. Only that kupdated (the kernel
version of caching disk access) is flushing more often.

> If you have lost several of your recordings thru inconsistencies
> in the filesystem I think you will think a bit different.

I use ext3. I have never lost any recording because of
power failures (HDD failure would a different story, of course).

> But of course, it's your choice.

CU,
Christian.



Home | Main Index | Thread Index