How much Swap is too much?

I have a HP-UX B.11.23 server with 16 gb of memory 84 gb of swap configured. I am being pushed to define more swap to try and get more Tuxedo domains to start. At what point do we have too much swap for the amount of memory? Thanks in advance.

If you ask me, you already have too much swap. You should increase memory or implement in a cluster architecture.

The old rule used to be 3 x memory and we have to get people to stop following that. The old rule was conceived when 64 Megs was a lot of memory and kernels allocated memory differently.

You're currently at 5.25 x memory. That's the highest ratio that I have ever seen.

Edit: Or did you make a typo? 48 GB of swap would be the old rule. You must have switched the digits!

I hear you, cheap boss, development environment. The HP-UX servers were dumped in my lap after our old admin quit.

You can have your boss read this thread.

The days of using swap were because memory was expensive. Memory is very cheap now. Money should not be the reason to sacrifice performance for a few dollars of memory.

I did not switch the digits, and I did not configure the machine. Here is the swapinfo.

(hostname-root: /:147)-> swapinfo -tam
             Mb      Mb      Mb   PCT  START/      Mb
TYPE      AVAIL    USED    FREE  USED   LIMIT RESERVE  PRI  NAME
dev       16544       0   16532    0%       0       -    1  /dev/vg00/lvol2
dev       65536       0   65515    0%       0       -    0  /dev/vg02/swap04
reserve       -   64153  -64153
memory    16382    3524   12858   22%
total     98462   67677   30752   69%       -       0    -

Sorry about missing the code tags.

Well, if you have enough memory and you are not disk swapping, why are you asking about increasing swap?

All the Tuxedo domains are not running right now. Earlier today I loaded up six of them and all of swap was being used and no more would start. The boss told me to add more swap so we can run 10 instances. At that time I explained swap was probably already too high. I wanted to verify I was not stating something that was not true. The boss mentioned about not caring about nanoseconds of performance. Thanks so much for all the feedback.

I assume your boss was joking. Swap can ad many milliseconds, not nanoseconds. In addition, thrashing due to swap can add seconds or more.

I recommend clustering:

Tuxedo Clustering

Thanks guys for taking the time to provide the feedback and Clustering information. I really do appreciate it and I will look into the Clustering and see if I can improve the environment. Thanks again.

Ive seen by minutes when it comes to sync GB...

Be sure swapmem_on is set to 0
Reduce your buffer cache since you are on SAN
If you are using VXSF of at fine tuning the latter...
But of course all this will not improve like adding an extra 16GB RAM...

Why 0? I would have suggested 1.

Anyway, this page says that swapmem_on is now ignored and the kernel will act as if it's on.