top-glance

[Andi Rigauer]

hi everybody,

I wanna raise a question to u experts. How reliable is the top? what are your experiences and which tools can be used instead of using the top (besides glance).

Thanks in advance
Andi

[Nobody's Hero · Accepted solution]

I have found top to be accurate. I like to use it to see the memory hogs running on my systems. Each process shows me the amount of memory it is using. Always check the total % for cpu usage. I had to install a top patch because even though the processors were 50% idle, the total still showed 100% idle. Load the latest top accum patch.

[doug mielke]

I find top to be accurate, and is a good tool for finding resource hogs. I've had trouble interpreting Glance output, however.

Documentation on just how Glace arrives at it's numbers seems lacking in some areas.
(i.e. %disk busy)

I'm a huge fan of sar, limited functionality, but it's been around so long, that documentation is easy to find, easy keeps history, and it's free.

[Lorenzo Facello]

Hi,
I suggest you to follow top'stat.
Some times infact glance doesn't match with top.
Maybe you have to install some glance's patches, (stating glance and hpux version)
L.

[Mark Grant]

top is good for a quick overview of what is happening but you can't beat "vmstat" and "iostat" for real data. They look ugly but as with so many things, the uglier, the more willing to please.

glance is a very handy tool too and I too, have a soft spot for "sar"

[John Carr_2]

top works very well and you dont have to pay for it like glance, although you can download a 60 day evaluation copy to try for yourself. As Mark said iostat and vmstat give the real figures and its probably what glance uses to give you a pretty output.

:-) John.

[Massimo Bianchi]

Hi,
in my experience top is very inaccurate, the higher the hpux the less reliable it is.

I trust glance metrics (if avaiable, since is a pay-product), and always trust sar and vmstat.

For monitoring disks performance you can also use iostat.

If you want to collect performance metrics, you can also think of getting mwa.


uptime can also be a good font for run-queue statistics

glance has many other options, like the syscall list, the memory region list, the open files list, otherwise you have to get many other products to have this information.

[Alex Ostapenko]

Dear Andi --

I don't use "top" because it's "officially not supported" or it's only "unofficially supported" by HP. There's no guarantee of its accuracy, problem fixes, or future support. You might as well bite the bullet and become familiar with some of the official products. Previous replies list those for you. "glance" is good for real-time data and has a lot of options. Use "?" for a list of options and "h" command for detailed help on each screen.

"measureware(mwa)" (which comes with Glance Plus Pak) is fantastic for historical data. This is the tool of choice of many HP Performance Analysts. You can easily dump data that's importable and graphable by MS-Excel. Or you can use Perfview (a separate product) to look at and analyze that data without having to import it into your PC. Both "glance" and "mwa" use the HP-UX measurement interface daemon (midaemon) that has hooks into the kernel for very detailed performance metrics.

The UNIX command-line tools are also very powerful. I've attached a sample daily report that I use for a number of our servers. The report rolls up hourly summaries of data sampled every 10 minutes. With the magic of Kornshell and awk, you can have nice "sar", "vmstat", "uptime", "swapinfo", "netstat -i", and "bdf" summaries. See my attachment. One word of caution, I don't trust sar's "avwait/avserv" fields, so I compute my own service time:
service-time = 10 * (%busy) / (r+w/s)

Happy performance tuning!

=:-) Alex