filecache_max: Will the usage (ever) reduce?

[jason_lee]

Hi,
Our server:
1) RX3600
2) HP-UX 11.31
3) 8GB RAM
4) filecache_max = 35%
5) filecache_min = default (5%).

I noticed from "kcusage" that the usage for filecache_max keep increasing, my question is will it reduce? Or it will just hit 100% when I don't reboot it after sometime?

Points for useful idea.
Jason.

[Patrick Wallek]

My understanding is that filecache_max is much like the dbc_max_pct of previous HP-UX versions. The usage of memory for file caching can increase up to the max percentage of memory configured (35% in your case). If you ever run into memory pressure, then the usage should decrease as needed to relieve that memory pressure.

[Dennis Handly]

>from "kcusage" that the usage for filecache_max keep increasing

I thought the values you get from kcusage(1m) will always increase, since its job is to track the maximum, not the current. But it seems to say it is current.
kcusage(1m) also says some interesting things about filecache_max. If near the limit, you may want to reduce:
The usage data for filecache_max may mean the opposite. ...

[jason_lee]

Dennis,
You said:
>since its job is to track the maximum, not >the current.

It should be the current.

>If near the limit, you may want to reduce:
How come? Shouldn't we increase it?

>The usage data for filecache_max may mean >the opposite. ...

Can you elaborate on this?

Well, what I wanted to know is, what's the optimum value that I should set for this filecache_max? Previously it is default (50%), now, I've changed it to 35%, running for a week, there is no problem at all.

[Dennis Handly]

>It should be the current.

(Yes, I corrected myself.)

>How come? Shouldn't we increase it?

Read kcusage(1m) where it discusses how filecache_max may be the opposite.

>I've changed it to 35%, running for a week, there is no problem at all.

This is what kcusage(1m) says. If you have it too big, it may be wasteful. That 35% may still be too big. What are you running on your system?

[jason_lee]

Dennis,
You said:

>This is what kcusage(1m) says. If you have >it too big, it may be wasteful. That 35% >may still be too big. What are you running >on your system?

I'll read the man page later. We are running:
i) Oracle 10g with DB mem footprint = 350MB
ii) OMU 8 with < 300 nodes
iii) NNM 7 with ET turned on

I believe 35% = 2.8GB, is huge, but I set this based on the max value this filecache_max has hit (from SMH's historical graph).

[Venkatesh BL]

If setting filecache_max to 35% does not create memory pressure on the system later, then you should be fine with it.

The actual consumption of the filecache memory would never go beyond the filecache_max value that was currently set on the system.

If you run into memory pressure situation (vhand consuming lots of CPU cycles), then you could consider change the filecache_max value.

[jason_lee]

Vent,
Thanks for your reply. So, we can say that, as long as vhand don't start consuming lots of CPU (can see it in the "top" output), then we are fine. I'm thinking of developing a OM monitor to check this.

[Venkatesh BL]

Yes, vhand will come into action when either the entire system or the file cache is under pressure. When that happens, you can see the cause for the pressure (could see the filecache consumption at the time). Based on the observation, you could either decrease or increase the filecache_max value.

Note that the system file cache usage reaching filecache_max is *not* necessarily a problem. It is really a problem if you hit the 'max' limit very often and your application suffers due to that.