Very slow load tran

[unknown]

Hi,

currently setting up a 'warm' database using transaction
logs from primary, but it's taking rather long time.
The dump file is loaded in matter of seconds, and after the
analysis can take more than 30 mins to start the redo, which
in turn can take 20 mins.
As strange as it can be, a 1,8 Gig dump file takes 50mins to
complete while a 3,2 Gig file takes only 15mins more,
meaning that the bigger the file, better the MB per minute.

Does the load tran, just like the load database, initializes
all the log devices after it loads the file onto syslogs ?

This is an ASE 12.5.3 with EBF 13325 runnig on Solaris 5.9
Database size is 1Tb, and it took 4 hours to load from full
dump, so I think it performs OK. Only the transaction log
loads are crawling.
Disks are Veritas managed, devices created with quick io,
box is fully dedicated to sybase, nothing besides the load
is running.

Any ideas ?

Thanks a lot

[A.M.]

Post by unknown
currently setting up a 'warm' database using transaction
logs from primary, but it's taking rather long time.
The dump file is loaded in matter of seconds, and after the
analysis can take more than 30 mins to start the redo, which
in turn can take 20 mins.
As strange as it can be, a 1,8 Gig dump file takes 50mins to
complete while a 3,2 Gig file takes only 15mins more,
meaning that the bigger the file, better the MB per minute.

Not quite. It depends on the nature of the transactions
and the amount of work that has to be done.

Post by unknown
Does the load tran, just like the load database, initializes
all the log devices after it loads the file onto syslogs ?

No, there's no reason why it should need to. Syslogs
works on a rotational basis (syslogs is cyclic - except
when it fills). I'm not sure where a load trans would
place the log - its either from the start or at the
current logptr. It doesn't really matter, I suppose.

Post by unknown
This is an ASE 12.5.3 with EBF 13325 runnig on Solaris 5.9
Database size is 1Tb, and it took 4 hours to load from full
dump, so I think it performs OK. Only the transaction log
loads are crawling.

A very small update statement can affect a lot of pages
(say a transaction updates all records in a large table
with a value). You'd need to look at the log closer.

Post by unknown
Disks are Veritas managed, devices created with quick io,
box is fully dedicated to sybase, nothing besides the load
is running.

You should really use raw devices. Striped is preferable.
Run sp_sysmon while the load tran is running and check
to see if it shows anything unusual.

-am © MMVI

[unknown]

True, but one interesting piece of info is that a second
database is beeing loaded with those same files and it's
taking 1/2, 1/3 of the times this one is.
Same sun host (Sun Fire V1280), 8 USparc IV+ CPUS and 32
Gigs of RAM.
Both hosts are sybase dedicated. Config files are the same,
so is the solaris patch level.
Only thing that changes is the disk arrays. EMC on one side
and IBM Sharks on the other.
Veritas VM version is slighty different, but that shouldn't
be an issue (so I think...).

[A.M.]

Post by unknown
True,

What is true? You are replying to me, aren't you?

Post by unknown
but one interesting piece of info is that a second
database is beeing loaded with those same files and it's
taking 1/2, 1/3 of the times this one is.

So the exact same dump files are being loaded twice?

Post by unknown
Same sun host (Sun Fire V1280), 8 USparc IV+ CPUS and 32
Gigs of RAM.
Both hosts are sybase dedicated. Config files are the same,
so is the solaris patch level.

Same? Both? Is it one machine or two? And, if two,
is one dumping and one loading or are both loading?

Post by unknown
Only thing that changes is the disk arrays. EMC on one side
and IBM Sharks on the other.

Which one is faster/slower?

Post by unknown
Veritas VM version is slighty different, but that shouldn't
be an issue (so I think...).

No.

-am © MMVI

[Ramon Vennik]

Post by unknown
True, but one interesting piece of info is that a second
database is beeing loaded with those same files and it's
taking 1/2, 1/3 of the times this one is.
Same sun host (Sun Fire V1280), 8 USparc IV+ CPUS and 32
Gigs of RAM.
Both hosts are sybase dedicated. Config files are the same,
so is the solaris patch level.
Only thing that changes is the disk arrays. EMC on one side
and IBM Sharks on the other.

physical disklayout is the same ? (stripes, interleave, etc etc ?)
both raw (of cooked ?)

Post by unknown
Veritas VM version is slighty different, but that shouldn't
be an issue (so I think...).

[Ramon Vennik]

Since you're using veritas using quickIO, you've been reading
http://eval.veritas.com/mktginfo/products/White_Papers/Storage_Server_Management/sf_sybase_40_perf_solaris.pdf

or not ?

Did you get your performance benefits using quick IO compared to raw IO ?

- Ramon

Post by unknown
Hi,
currently setting up a 'warm' database using transaction
logs from primary, but it's taking rather long time.
The dump file is loaded in matter of seconds, and after the
analysis can take more than 30 mins to start the redo, which
in turn can take 20 mins.
As strange as it can be, a 1,8 Gig dump file takes 50mins to
complete while a 3,2 Gig file takes only 15mins more,
meaning that the bigger the file, better the MB per minute.
Does the load tran, just like the load database, initializes
all the log devices after it loads the file onto syslogs ?
This is an ASE 12.5.3 with EBF 13325 runnig on Solaris 5.9
Database size is 1Tb, and it took 4 hours to load from full
dump, so I think it performs OK. Only the transaction log
loads are crawling.
Disks are Veritas managed, devices created with quick io,
box is fully dedicated to sybase, nothing besides the load
is running.
Any ideas ?
Thanks a lot

[t***@none]

Hi there,

We may have a similar problem, but do not know where to start. SUN etc.
Have you solved your issue yet, and if so, what was the solution?

DBA...

Post by unknown
Hi,
currently setting up a 'warm' database using transaction
logs from primary, but it's taking rather long time.
The dump file is loaded in matter of seconds, and after the
analysis can take more than 30 mins to start the redo, which
in turn can take 20 mins.
As strange as it can be, a 1,8 Gig dump file takes 50mins to
complete while a 3,2 Gig file takes only 15mins more,
meaning that the bigger the file, better the MB per minute.
Does the load tran, just like the load database, initializes
all the log devices after it loads the file onto syslogs ?
This is an ASE 12.5.3 with EBF 13325 runnig on Solaris 5.9
Database size is 1Tb, and it took 4 hours to load from full
dump, so I think it performs OK. Only the transaction log
loads are crawling.
Disks are Veritas managed, devices created with quick io,
box is fully dedicated to sybase, nothing besides the load
is running.
Any ideas ?
Thanks a lot