Showing results for 
Search instead for 
Did you mean: 

Need to be able to justify request for more memory.

Go to solution
Jack C. Mahaffey
Super Advisor

Need to be able to justify request for more memory.

Need ideas on how to convince management to purchase additional memory for my servers. May also need additional cpus. What metrics do I need to prepare? I have glance on my system. I'm running two Oracle databases with about 300mb SGA each. There are upwards of 2000 processes on the system. My K380 has 2gb memory. Server also uses a Model E 12H Sure Store disk array with 12 drives for the application and database.

swapinfo -mt shows the following on the k380
dev 1000 439 561 44% 0 - 1 /dev/vgroot/swap
dev 2048 442 1606 22% 0 - 1 /dev/vg01/lvol1
dev 2000 444 1556 22% 0 - 1 /dev/vg01/lvswap2
dev 500 439 61 88% 0 - 1 /dev/vg01/lvswap3
reserve - 2102 -2102
memory 1537 442 1095 29%
total 7085 4308 2777 61% - 0 -

Here's a snapshot of top:
System: server1 Tue Feb 19 10:42:22 2002
Load averages: 3.13, 3.43, 3.50
1972 processes: 1930 sleeping, 38 running
Cpu states:
0 3.26 45.5% 0.0% 54.5% 0.0% 0.0% 0.0% 0.0% 0.0%
1 2.99 40.0% 0.0% 60.0% 0.0% 0.0% 0.0% 0.0% 0.0%
--- ---- ----- ----- ----- ----- ----- ----- ----- -----
avg 3.13 42.6% 0.0% 57.4% 0.0% 0.0% 0.0% 0.0% 0.0%

Memory: 925596K (181788K) real, 2135976K (309880K) virtual, 21712K free Page# 1

0 ? 0 root 127 20 0K 0K sleep 3617:24 5.89 5.87 swapp
0 ? 21799 root 152 20 340K 176K sleep 426:54 5.27 5.26 sh
0 ? 26853 oracle 154 20 10716K 1032K sleep 8:28 4.95 4.94 oracl
0 ? 3292 cbs1mgr 199 20 5652K 2860K run 0:00 16.86 4.37 MxUpd
0 ? 26810 cbs1mgr 158 20 8468K 5716K sleep 7:11 3.81 3.80 MxUpd
0 ? 2617 root -16 20 6128K 5448K run 2565:24 3.73 3.73 midae
0 pty/ttyu5 20258 llk4 154 20 13808K 9736K sleep 0:10 3.52 3.51 MxMai
1 pty/ttywe 13181 spd 154 24 5712K 1604K sleep 0:04 2.49 2.48 MxMai

I can't remember from class how much memory per Oracle user should be available. Looks like there currently are over 330 Oracle sessions.

Any suggestions will appreciated. I'd hate to bring in outside help for them just to say there are too many processes. I'd like to provide some convincing metrics. I'd like to be able to make a reasonable recommendation on how much more memory is necessary.

Patrick Wallek
Honored Contributor

Re: Need to be able to justify request for more memory.

From the looks of your swapinfo output, you appear to already be short on memory, as you are using swap space. Keep an eye on the memory usage in Glance as well. If it stays up at 100% for long stretches, then you do have memory pressure and more is definitely needed. Anytime you swap you will be slowed down significantly since you have to write to / from the disks, a slow operation compared to accessing the info. directly from RAM.

It also appears from your sample output that you can use more CPUs as well. It looked like your idle % on the CPUs is 0, so you also have a CPU bottleneck.

I don't recall how much memory a K380 can handle, but I would at least double it (go to 4GB) and an additional CPU board, if there is room, would be good as well.

If you really want good performance, an upgrade to a 4-way L2000 with about 8GB of RAM would be a great investment and your maintenance costs would probably be less, although the upfront cost, even on the re-marketed market, will be more.
A. Clay Stephenson
Acclaimed Contributor

Re: Need to be able to justify request for more memory.

Hi Jack:

The most important metric is the page out rate; if it is high then you definitely need more memory. In you have Glance and MeasureWare then there is a good chance that you have PerfView. I find that it is an excellent tool for making the kind of graphs that you need. Management understands 'Red bad; Green good'. I will say that 300MB SGA's are rather small for Oracle instance nowadays and 2GB's goes very fast.

I suspect that your other killer is the configuration of your 12H; I'll bet that most of it is allocated and that you haven't spread the I/O over both external busses by stripping each logical volume.

If it ain't broke, I can fix that.
harry d brown jr
Honored Contributor

Re: Need to be able to justify request for more memory.


If you are running 10.20, there is a max on memory, and I believe it's around 3.75gb. Are the users complaining that the system is too slow? If they are, then either an upgrade in memory and CPU's should be looked into. And I would look into the refurbished market for this, and HP is in this game also.

live free or die
Live Free or Die
harry d brown jr
Honored Contributor

Re: Need to be able to justify request for more memory.


I forgot to add: If you go for more CPU's, go for 2 more, because 1 more will not give you a 33% performance boost, it'll give you more like 20%, at best! You might even consider going to 11i on an N-class server, as they stomp K's!

live free or die
Live Free or Die
Mark Greene_1
Honored Contributor

Re: Need to be able to justify request for more memory.

Your oracle var should be able to provide you with a standard recommended config based on the size of the db, the number of users, and the type of activity (transaction processing, mostly queries, mostly archival, etc.).

The fact that top is reporting your system as 0% idle is an indicator that something needs upgraded, and you should be able to get corroboration on that from your hardware var. However, I would suggest first looking at upgrading your disk subsystem before adding cpu's. What does vmstat show for I/O wait? The through-put on the 12H arrays is only 20mbs per channel; if you upgrade to an sc10 or fc10 you can quadruple that. If you move to a fibre rack (sorry, don't recall the model numbers), you can get over 100mbs per channel.

the future will be a lot like now, only later