How can I see what OS buffer cache this system is using?

[Eric Antunes]

I have set dbc_min_pct and dbc_max_pct, respectively, to 5 and 10 and now I want to see if this is the correct interval, how can I monitor the Dynamic Buffer Cache??

Thanks,

Eric Antunes

[Pete Randall · Accepted solution]

Eric,

You can use sar -b and monitor the %rcache figure. It should be well into the nineties.


Pete

[Eric Antunes]

Pete,

From this sample above, should I change any value from the interval?

11:36:52 bread/s lread/s %rcache bwrit/s lwrit/s %wcache pread/s pwrit/s
11:36:56 89 274 68 58 95 39 0 0
11:37:00 58 158 63 67 50 0 0 0
11:37:04 154 354 57 93 105 11 0 0
11:37:08 145 892 84 111 122 9 0 0
11:37:12 10 490 98 113 140 19 0 0
11:37:16 58 362 84 26 23 0 0 0
11:37:20 34 654 95 65 135 52 0 0
11:37:24 26 299 91 6 15 63 0 0
11:37:28 31 224 86 4 3 0 0 0
11:37:32 6 591 99 4 18 80 0 0
11:37:36 9 762 99 48 110 57 0 0
11:37:40 0 32 98 10 12 19 0 0
11:37:44 7 12 38 4 5 30 0 0
11:37:48 18 1673 99 22 22 0 0 0
11:37:52 6 169 96 10 6 0 0 0
11:37:56 2 541 100 29 62 52 0 0
11:38:00 0 4 100 4 4 0 0 0
11:38:04 0 23 100 3 5 35 0 0
11:38:08 0 588 100 10 10 3 0 0
11:38:12 0 6 100 4 5 16 0 0

Average 33 405 92 34 47 27 0 0

Thank you,

Eric

[Fred Ruffet]

It should be good for read. For writes, it is a bit low, but it has not the same impact...

Note that read percentage is something exponential : It will be easy to go from 10 to 20 percent by increasing cache size, but to increase from 98 to 99 percent, you will need to buy trucks of memory.

Regarding write, wait for other advices, I can not be sure of a conclusion.

Regards,

Fred

[Pete Randall]

Trying to tune to increase write cache is pointless. Unless you're doing a lot of sequential writes, it is highly unlikely that a particular write request is going to be satisfied by data in the cache. Reads, on the other hand, tend to be more sequential and more easily satisfied by data in cache.

From your data, the read percentage occasionally looks a little low. That may or may not indicate a lack of cache. You might want to try increasing the cache and monitoring to see if things improve.


Pete

[Eric Antunes]

Fred,

Already a king? :)

Pete,

I'll increase the interval to 10-15 and see if I get better results...

Thanks,

Eric

[Fred Ruffet]

Yes :) I'm keeping my old magic wand and wizard hat for you. This won't be long now.

Something we didn't ask in this thread : How many RAM do you have on this machine ? What kind of app is running there ?

Regards,

Fred

[Eric Antunes]

I appreciate that! I just don't want to be more than I deserve so I'm going slowly to reach the wizard level by the end of the year.

The server has a 1.5 Gb RAM and I'm running Oracle E-Business Suite 11.0.3 and RDBMS 8.0.5 on it.

Kindest Regards,

Eric

[Fred Ruffet]

I know from preceding posts you don't use buffer for the DB, so I imagine, regarding your sar results, Business suites does lots of IO (I don't know exactly what it does). But I also think it uses memory. And so do your Oracle DB. With all those elements, I think you won't be able to grow your cache more without swaping. And I come to the same conclusion you have some threads before :
You're short on RAM.

Regarding all of that, I maintain that your ratios are not so bad, and first thing you should do is to reclaim RAM for this server.

What gives sar -w (swap info) ?

Regards,

Fred

[Eric Antunes]

Fred,

Yes I use Buffer Cache for the DB: 3000 db_block_buffers=24Mb.

Not at all, E-Business Suite doesn't do lots of IO: custumizations (that's the functionalities you add) do! That's why I keep telling people that for Oracle environments SQL tuning, good tablespace management (few great extents), little chaining, etc... is where you get great performance improvements!

We are closing the month: that's when the system gets more busy but here, see this (No swaping at all):

# sar -w 4 100
...

15:26:09 swpin/s bswin/s swpot/s bswot/s pswch/s
15:26:13 0.00 0.0 0.00 0.0 259
15:26:17 0.00 0.0 0.00 0.0 311
15:26:21 0.00 0.0 0.00 0.0 259
15:26:25 0.00 0.0 0.00 0.0 234
15:26:29 0.00 0.0 0.00 0.0 287
15:26:33 0.00 0.0 0.00 0.0 218
15:26:37 0.00 0.0 0.00 0.0 242
15:26:41 0.00 0.0 0.00 0.0 223
15:26:45 0.00 0.0 0.00 0.0 278
15:26:49 0.00 0.0 0.00 0.0 288
15:26:53 0.00 0.0 0.00 0.0 404
15:26:57 0.00 0.0 0.00 0.0 328
15:27:01 0.00 0.0 0.00 0.0 831
15:27:05 0.00 0.0 0.00 0.0 280
15:27:09 0.00 0.0 0.00 0.0 304
15:27:13 0.00 0.0 0.00 0.0 284
15:27:17 0.00 0.0 0.00 0.0 291
...
15:31:49 0.00 0.0 0.00 0.0 272
15:31:53 0.00 0.0 0.00 0.0 546
15:31:57 0.00 0.0 0.00 0.0 536
15:32:01 0.00 0.0 0.00 0.0 416
15:32:05 0.00 0.0 0.00 0.0 430
15:32:09 0.00 0.0 0.00 0.0 244
15:32:13 0.00 0.0 0.00 0.0 384
15:32:17 0.00 0.0 0.00 0.0 268
15:32:21 0.00 0.0 0.00 0.0 450
15:32:25 0.00 0.0 0.00 0.0 473
15:32:29 0.00 0.0 0.00 0.0 294
15:32:33 0.00 0.0 0.00 0.0 307
15:32:37 0.00 0.0 0.00 0.0 291
15:32:41 0.00 0.0 0.00 0.0 392
15:32:45 0.00 0.0 0.00 0.0 368
15:32:49 0.00 0.0 0.00 0.0 349

Average 0.00 0.0 0.00 0.0 338

Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 1024 0 1024 0% 0 - 1 /dev/vg00/lvol2
dev 1024 0 1024 0% 0 - 1 /dev/vg00/lvol9
reserve - 522 -522
memory 1061 93 968 9%
total 3109 615 2494 20% - 0 -

Best Regards,

Eric

[Fred Ruffet]

I know it's a bit confusing, but I was talking about System's cache (dynamic buffer cache). I believe you mount FS with convosync and mincache options.

Regards,

Fred

[Eric Antunes]

Fred,

I just use delaylog,nodatainlog (don't have Online JFS to use convosync and mincache options.)

Regards,

Eric

[Fred Ruffet]

So (It's just my feeling) I thing you won't have gain on that %rchache ratio.

As I already said, due to double-cache overhead, Oracle will always ask for more and fill OS buffer. So your cache will never be enough.

Once again, it's just a feeling, and you need to test new values. You can as long as your machine do not seem to swap. But go slowly on increments as long as it is prod, and if this parameters are too high, cache will grow and grow and grow...

Regards,

Fred

[Emil Velez]

Since this is a HPUX system do you have glance plus pak which includes the openview performance agent (AKA measureware) that collects hundreds of metrics a few of which include the current size of buffer cache and you can use that collected data to generate reports on its size, cache hit rate, read hit and write hit. This data collection (if installed) is ongoing with little overhead and I run into many people who have this product loaded and running (with the MCOE or Enterprise OE) but they do not realize the wealth of data that has been collected.

[Eric Antunes]

Thanks for all replies.