Share This:

How much RAM do you need?


I am old enough to recount:

  • The first computer I touched at school had 1 kB
  • The computer my Dad bought me had (what seemed to be) a gigantic 64 kB
  • The first Sun workstation I used sported 32 MB
  • For numerical modelling, I used a supremely expensive HP workstation, which had 384 MB

Now, you'd think nothing if your desktop machine came with 8 or 16 GB.


So, how much RAM do you need on your ADDM/BMC Discovery appliances?


We have documented some sizing guidelines, which suggest 16 to 32 GB for a consolidation appliance. What is not so clear though, is if any improvement will be made from having more, and if so how much. Clearly, the system needs a minimum amount or RAM for running the OS and all the appliance stack components comfortably, and without frequent page ins/outs. Beyond that, additional RAM can be utilized for caching IO to the datastore files, and this is where you may see additional benefits.


I ran some IOzone tests to show how random reads and write performance changes for different file sizes and appliance RAM (only up to 16 GB, sorry. Donations for my memory upgrade fund gratefully accepted).


You can see that for small file sizes the performance doesn't vary much with RAM - as long as the entire file can be cached, that's as good as you are going to get. However, once the file size gets too big, performance drops off a cliff. You can see that adding more RAM to the system pushes that point higher.


The lesson is - if you have large datastore files, you will probably get better datastore performance by adding RAM over an above the traditional recommended minimums. A real customer example was seen, while monitoring doomed node deletion rates:


  • With 46 GB RAM, the deletion rate was about 4 million deletes/day
  • After upgrading to 144 GB RAM, the rate increased to about 17 million deletes/day.



(Test details: single 10.0.03 appliance on VMware ESXi 5.0.0 on HP DL380G5, Hardware RAID 5 local disks. 4 runs for each configuration)


Join the Customer Support Community and give us feedback if our efforts in Communities are helpful, and how we can better serve you.