SCO OPENSERVER 5 ON A VM IN INFRASTRUCTURE 3 ESX SYSTEM

[morenos]

I've moved an Unix SCO Operating System's image (Rel.5.0.6) from a Pentium3 hardware to a new generation hardware's Server. This virtual machine works correctly, it's very fast in disk access, but it's very slow in memory calculations. I give you a simple example.

I've written a simple script that performs a loop of 100000 cicles and increase a variable by 1 in every cycle. At the end print the result.

I run this script on the physical machine (Pentium 3 1.2Ghz) and it finishes after 2'18". The same script on the virtual machine (with Xion's CPU 1.60GHz hardware) finishes after 16'49". If I try the same script in a Linux's virtual machine on the same vmware's server, it finish after 0'42".

During the execution of this script, if I try the sar unix command, I find that the CPU's queue is always full (100%).

Have you any idea how can I resolve this problem?

Thanks a lot.

This is the simple shell script (/tmp/a):

\##########################

i=0

while \[ $i -lt 100000 ]

do

i=`expr $i \+ 1`

done

echo $i

\##########################

I've tried to create a new vm with SCO Unix Openserver operating system version 5.0.7, but after installation I can't start the system because it show the message:

out of low memory

Any suggestions?

[larstr]

Keep in ming that you're forking 100000 times in your script. This means that you're doing a lot of kernel calls, and kernel calls are not as fast as user calls, independent of the guest OS.

Try rewriting that script in perl, awk or similar where you don't have to fork as much.

Also see the following thread:

http://www.vmware.com/community/thread.jspa?messageID=673024

SCO also has a kb article on installing OSR in VMware.

Lars

[morenos]

Hi Lars,

thanks for your hints.

My script is only a simple performance's test on the vm with SCO Operating System.

I tried some Unix's applications and verified that they were slower that on old physical server. I verified that the hd's access were very fast (because the server esx has SAS's hds), but the calculation's time were very very slow. I notice this also when I run a du command (for instance du /usr) to calculate the directory occupation on disk. If I run the sar command while the command is running, I see this result:

07:03:39 %usr &sys &wio %idle

07:03:40 0 97 3 0

07:03:42 0 95 5 0

07:03:43 0 97 3 0

07:03:44 0 95 5 0

07:03:45 0 99 1 0

Average 0 97 3 0

As you can see, the bottleneck is cpu's access.

The same tests, on a linux vm, are very very faster. For this reason I suspect that the problem is Unix SCO's operating system that doesn't work fine with the virtual cpu of vmware. Can it depend that I haven't vm tools for this operating system? Or the cause is the hyper-threading technology of my esx server? It works better with SCO 5.0.7 version?

If someone has any ideas I'd be very happy because I have several Unix's physical server to migrate on vmware (our servers and customer's servers).

Thanks.

Moreno.

[continuum]

Which guestOS do you use ?

[morenos]

The Operating System of the virtual machine is Unix SCO Rel.5.0.6. I configured the vm with Guest OS: Other.

[larstr]

Did you give the VM one or two cpus?

Is your kernel an SMP kernel?

How is the %READY figures of this VM?

Lars

[morenos]

Hi Lars,

I gived one CPU to this VM. The kernel hasn't SMP module (it costs a lot).

What do you mean for %READY figures?

Moreno.

[larstr]

Moreno,

The %READY figures tell you how much cpu your guest didn't get due to contention. Traditionally is has been recommended that the %READY

values are below 5%[/url] in order to have good performance of your VM. However, we're not getting all values in % in VIC, so you'll have to compute

it of the sampling rate[/url].

There are some documents on the values here:

http://www.vmware.com/pdf/esx3_ready_time.pdf

http://www.vmware.com/pdf/vi3_monitoring_statistics_note.pdf

In VIC you typically see the cpu usage+ready like this:

I'm also working on creating some performance charts for VI3 where you amongst others get the READY figures in %.

Lars

[morenos]

OK, I verified the %READY on my VIC. You can see it on the following link (excuse me, but I didnt' found another method for show it). As you can see, I performed my shell script between 15:12 and 15:29 o'clock.

[larstr]

During that period your %READY is 3-4%, and that's within the previous mentioned limit and should not cause you your bad performance.

How are the scripts in the other thread[/url] running? How long time do they take on your system? Xeon 1.6 also comes in several flavors. Is that a 4 years old cpu, or is it a new one? (P3 Xeon, P4 Xeon or Core2 Xeon? DP or MP?)

Regarding SCO Unix, I haven't really used it in quite a few years, and there aren't too many questions regarding it in these forums either, so I really don't how it performs in a VMware environment. It's an unsupported guest OS, and there are no VMware tools available. If possible, I would plan to migrate away from SCO to a supported platform (such as a supported linux or windows platform).

Lars

[morenos]

I tried this script on my SCO's VM and it needs 16'; the same script on a Linux's VM (on the same esx server) needs 4' (not 0'52" as I wrote initially, sorry). I tried the same script on the console of my esx server and here it needs only 1'51".

The server has 1 cpu Xeon 5000 Dual Core DP processor, bought 1 year ago (server FSC FUJITSU-SIEMENS RX300S3).

I know that SCO isn't supported, by I cannot migrate these applications on others operating systems (they are old applications). I hope to find a solution on vmware environment because I want to eliminate the old hardware servers.

[morenos]

With vmware 4.x, SCO Openserver 5 is a operating system supported from vmware. I tried to move some server to this infrastructure and they work very well.