Re: Archive process too slow - please help

From: Bradley W Howerter <howerter_at_world.std.com>
Date: Fri, 6 Jan 1995 14:28:54 GMT
Message-ID: <D1zMw6.13x_at_world.std.com>


howerter_at_world.std.com (Bradley W Howerter) writes:

>My archive background process cannot keep up with the database writer.
>Even though we have more than 1Gbyte of redologs, I end up getting
>the error "All online redologs needed archiving" because the
>archive process is too slow, and it is still working on the
>redolog file that is needed by the database writer.
 

>We figure we could work around this by adding even more disk space
>for the redologs, but I'm wondering if there are any other alternatives.
 

>This is Oracle V7.1.3 on Sequent PTX.

It is me again. Most of you asked for more details. Here they are:

First of all, let me clarify the problem. We only have problems with the archive background process at night, when we are running our batch mrp program. It is VERY heavy I/O. I am trying to convince management that it should be redesigned to do less I/O. But in the meantime, I'm expected to try to tune for it. When mrp stops at 7 in the morning, the archive process catches up.

We have 4 logfiles, 260 Mbytes each. We used to have 30 10Mbyte logs, but on the advice of a consultant we drastically increased the size of these. This eliminated the problem until recently. We think the reason the problem has cropped up again is that mrp was tuned. This causes it to do even MORE I/O.

At night we switch redo logs every 20 minutes. During the day we switch one time or zero times. Rarely two times.

Yes, we archive to a disk, not tape. No, the disk isn't filling up, but you are right that when it does, it is hard to recover. (We create a symbolic link to a non existent directory to keep oracle from writing to its next archive file when this happensotherwise,  the copy from disk to tape is VERY slow.)

No, we are not striping. No, there are no conflicts with other datafiles and the redologs. We have two disks reserved for redologs. There is also no conflict with archive logs disk drive. The redo logs are not yet raw, nor is the database, so we have multiple database writers. We intend to make the redo logs raw this weekend. I can't think of a good way to make the archived files raw.

Yes, we only switch when the log file fills up. We have log_archive_buffer_size=256, and
log_archive_buffers=5.

Thanks to all who responded for your concern and your help, and I'll keep you posted if we find a solution. Received on Fri Jan 06 1995 - 15:28:54 CET

Original text of this message