What does it mean "Fuzzy Checkpoint"?
Posted in 2000
Topics: Logging & Checkpoints, Versions, Editions & End-of-Life
Hi all,
I'm moving from IDS 7.30 to IDS2000 and still learning with it. There is
some new term as "Fuzzy Checkpoint" in online.log that I don't see in IDS
7.30 for example:
17:00:14 Fuzzy Checkpoint Completed: duration was 0 seconds, 77 buffersnot flushed.
17:00:14 Checkpoint loguniq 299, logpos 0x59f9c2a4
Is there something different between IDS 7 and IDS2000 in checkpoint area or
not?
Thanks in advance for any advise.
DH.
IDS 2000 introduces the concept of a "fuzzy checkpoint", whereby not all
dirty pages are written back to disk at checkpoint time. A change to the
recovery logic, involving the use of logical logs, has been made to
facilitate this. The advantage is that pages may be retained in cache for
longer, thus increasing the cache rates.
I think it may be possible to disable it and have "real" checkpoints if
required. To be honest, though, in my experience it's pretty much
transparent to sys admins/DBAs.
DH <huyv@usa.net> wrote in message news:975033368.383729@news.aei.ca...
> Hi all,
>
> I'm moving from IDS 7.30 to IDS2000 and still learning with it. There is
> some new term as "Fuzzy Checkpoint" in online.log that I don't see in IDS
> 7.30 for example:
>
> 17:00:14 Fuzzy Checkpoint Completed: duration was 0 seconds, 77 buffers> not flushed.
> 17:00:14 Checkpoint loguniq 299, logpos 0x59f9c2a4
>
> Is there something different between IDS 7 and IDS2000 in checkpoint area
or
> not?
>
> Thanks in advance for any advise.
>
> DH.
>
>
>
Systems that have too many dirty pages at checkpoint time, are very slow to
checkpoint, even if you set LRU_MIN_DIRTY/LRU_MAX_DIRTY to very low
numbers. So, fuzzy checkpoints were created to solve the "blocking" of
user activity during a checkpoint. If you haven't experienced this
problem, you may not see the advantage of fuzzy checkpoints.
it is possible to get the full sync checkpoints as the default, if you set
an environment variable -- but I don't know how its spelled. check with
Informix.
Neil Truby wrote in message <8vm55e$t01$1@taliesin2.netcom.net.uk>...
>IDS 2000 introduces the concept of a "fuzzy checkpoint", whereby not all
>dirty pages are written back to disk at checkpoint time. A change to the
>recovery logic, involving the use of logical logs, has been made to
>facilitate this. The advantage is that pages may be retained in cache for
>longer, thus increasing the cache rates.
>
>I think it may be possible to disable it and have "real" checkpoints if
>required. To be honest, though, in my experience it's pretty much
>transparent to sys admins/DBAs.
>
>DH <huyv@usa.net> wrote in message news:975033368.383729@news.aei.ca...
>> Hi all,
>>
>> I'm moving from IDS 7.30 to IDS2000 and still learning with it. There is
>> some new term as "Fuzzy Checkpoint" in online.log that I don't see in IDS
>> 7.30 for example:
>>
>> 17:00:14 Fuzzy Checkpoint Completed: duration was 0 seconds, 77 buffers>> not flushed.
>> 17:00:14 Checkpoint loguniq 299, logpos 0x59f9c2a4
>>
>> Is there something different between IDS 7 and IDS2000 in checkpoint area
>or
>> not?
>>
>> Thanks in advance for any advise.
>>
>> DH.
>>
>>
>>
>
>
DH wrote:
>
> Hi all,
>
> I'm moving from IDS 7.30 to IDS2000 and still learning with it. There is
> some new term as "Fuzzy Checkpoint" in online.log that I don't see in IDS
> 7.30 for example:
>
> 17:00:14 Fuzzy Checkpoint Completed: duration was 0 seconds, 77 buffers> not flushed.
> 17:00:14 Checkpoint loguniq 299, logpos 0x59f9c2a4
>
> Is there something different between IDS 7 and IDS2000 in checkpoint area or
> not?
Yes. IDS.2000 completes the implementation of a new checkpointing method
called Fuzzy Checkpoints. They are explained in full in the 9.2x version
of the Administrator's Guide. Quickly, a fuzzy checkpoint does NOT flush
all dirty buffers as earlier versions did and does not create physical log
records for MOST dirty pages. It relies more heavily on the logical log
for recovery. The upshot is that checkpoints happen more quickly and the
reliance of the engine on CHUNK writes is reduced significantly meaning
that you can increase your LRU_MAX/MIN_DIRTY parameters without increasing
checkpoint times noticably and improve write cache performance and reduce
the IO load on the system.
Fuzzy checkpoints were partially, and silently, implemented in 7.30 and
7.31 but there was an undocumented variable needed to enable most of the
functionality there. In 9.2x Fuzzy Checkpointing is the default and a
documented variable can be used to disable it and restore the older Complete
Checkpoint code.
Art S. Kagel
Joanne Dority wrote:
>
> Systems that have too many dirty pages at checkpoint time, are very slow to
> checkpoint, even if you set LRU_MIN_DIRTY/LRU_MAX_DIRTY to very low
> numbers. So, fuzzy checkpoints were created to solve the "blocking" of
> user activity during a checkpoint. If you haven't experienced this
> problem, you may not see the advantage of fuzzy checkpoints.
>
> it is possible to get the full sync checkpoints as the default, if you set
> an environment variable -- but I don't know how its spelled. check with
> Informix.
onconfig parameter -
NOFUZZYCKPT 1
>
> Neil Truby wrote in message <8vm55e$t01$1@taliesin2.netcom.net.uk>...
> >IDS 2000 introduces the concept of a "fuzzy checkpoint", whereby not all
> >dirty pages are written back to disk at checkpoint time. A change to the
> >recovery logic, involving the use of logical logs, has been made to
> >facilitate this. The advantage is that pages may be retained in cache for
> >longer, thus increasing the cache rates.
> >
> >I think it may be possible to disable it and have "real" checkpoints if
> >required. To be honest, though, in my experience it's pretty much
> >transparent to sys admins/DBAs.
> >
> >DH <huyv@usa.net> wrote in message news:975033368.383729@news.aei.ca...
> >> Hi all,
> >>
> >> I'm moving from IDS 7.30 to IDS2000 and still learning with it. There is
> >> some new term as "Fuzzy Checkpoint" in online.log that I don't see in IDS
> >> 7.30 for example:
> >>
> >> 17:00:14 Fuzzy Checkpoint Completed: duration was 0 seconds, 77 buffers> >> not flushed.
> >> 17:00:14 Checkpoint loguniq 299, logpos 0x59f9c2a4
> >>
> >> Is there something different between IDS 7 and IDS2000 in checkpoint area
> >or
> >> not?
> >>
> >> Thanks in advance for any advise.
> >>
> >> DH.
> >>
> >>
> >>
> >
> >