HP-UX Health Status

[bernsceb]

Hi all,

I recently issued a report on the systems HDDs using the sar files. Most of the data available is quite straight forward and I got the hang of it. My issue is to indentify if the values which I have under 'r+w/s', 'blks/s', 'avwait', 'avserv' are good or alarming. What I need is for someone to guide me and tell how these values need to be in order to determine if the system is performing good or not.

Many thanks
Bernard

[Tim Nelson]

IMHO, the most valuable stat is avserv.

Depending on what type of disk is being reported on here would be my guidelines ( only an opinion)

Locally attach single disk ( like an internal drive ) aveserv < 20 is ok


SAN/Array based storage (depends on type of storage behind SAN )

aveserv 1-5 excellent
aveserv 6-12 ok
aveserv > 12 might have a problem
aveserv > 20 you really have a problem

Some of this has to be taken as only a possible indication of more investigation.

i.e. sparse small block randomly on an idle disk will typically show poor service times. This typically occurs on internal OS disks that have little utility but a small request here and there requires the heads to possibly move from one end of the platter to another this is where making note of r+w/s. Little activity sometimes equals little response.

Another example: a small block random request to storage array currently processing other large block sequential IO will seldom experience excellent response times ( this might be where you look at blks/s or queue lenth in conjunction).

Take all of these into consideration with %utility. If the storage is not busy then believe it nor not the service time may be higher (sometimes). Kinda like going from dead stop to 60mph. Takes a couple seconds to get up to speed. Nothing in cache, nothing in read-ahead, got to go all the way there to get it.

[bernsceb]

Many thanks for your help, this was very helpfull. The following are the results of my two machines. Both machines are quite powerfull and are set with two logical arrays (stripe), as far as I can say I think that they are currently working at almost optimum conditions.

c0t2d0 10.10 0.50 16 133 0.00 6.69
c3t2d0 5.05 0.50 8 101 0.00 5.91
c0t0d0 3.03 0.50 6 18 0.00 4.91
c3t0d0 1.01 0.50 3 6 0.00 4.71
c0t0d0 6.00 0.50 9 96 0.00 15.31
c3t0d0 2.00 0.50 6 84 0.00 10.14

[Steven E. Protter]

Shalom,

These numbers are merely a guide.

It is possible for these numbers to get very distorted, yet the system functions well.

The way I measure system health, besides these useful reports is response time. If possible, I use the application the server runs and check it out once in a while.

Also user reports of slow response are critical in initiating the process of running special reports and beginning the diagnosis process.

SEP