R82 using more swap than R81.20

[Don_Paterson]

Anyone know why an R82 SMS would use more swap than an R81.20 SMS?

Both have 8GB RAM.

The R82 SMS has 8 CPU cores and the R81.20 SMS has 4

Both are on Hyper-V (Build: 20348-10.0-3-0.3692)

R81.20 is unpatched (take 631) and R82 is JHFA Take 33

R81.20 uses 70M of swp and R82 uses up to 2G when doing nothing (idle) - and after 1 hour of uptime.

SmartConsole is connected but not used.

[Vincent_Bacher]

Is this a lab setup or why did you just assign 8GB of RAM?
Our two main MDS systems running R82 show 0MB and 10MB swap usage respectively -
but then again, we run them with slightly more RAM than 8GB 😉

[Don_Paterson]

😄

Yes, it is a lab setup and RAM is a little short in that lab.

I have a separate VMware Workstation lab and an R82 SMS with 16GB RAM also uses zero swap.

I am looking to see if there is a difference between the two versions that is making the difference.

Linux 3.10 and 4.18 is obviously significant. 

The mem used on all of them is around 4.5G.

[the_rock]

My colleague is upgrading eve-ng, so I will check once its up later on. Personally, did not see any difference with R81.20, but will confirm in few hours. Where did you verify this? cpview output, I assume?

[Don_Paterson]

htop and free -h

I found this so far:

https://access.redhat.com/solutions/7042476

Resolution

• Changes in virtual memory (VM) management between RHEL 7 and 8 may cause different patterns of swap usage even under similar workloads and tunings
• Swap usage or a trend towards 100% swap used is not necessarily indicative of memory pressure.
• Recommend monitoring other statistics in order to determine if the system is under memory pressure (see Diagnostic Steps)

I wonder if the changes in swap behaviour can affect performance in some cases where, for example, the storage of a lab is not super fast and swap usage exposes that.

Note:

R81.20 is RHEL 7 and R82 is RHEL 8

[Vincent_Bacher]

I think that even if there are changes in swap usage, this should not be noticeable in the lab. However, if a lot of swap is needed in production and the disks are extremely slow, then this is obviously not good.

[Don_Paterson]

I have indeed not seen a difference in many tests. However, there do seem to be some lagging issues during the busier times of the day when more users are using the shared lab platform.

Can't quite put my finger on it yet.

In one case the swp went to 4G but that was on a 6GB RAM SMS. That's before the upgrade to 8G installed into the VM.

[Vincent_Bacher]

Nevertheless, I would never configure less than 16GB, even in the lab.

[the_rock]

100%, thats safest bet, in my opinion. I guess 12 GB could work, but probably not enough in a long run.

[Don_Paterson]

With you 100% 

Sometimes we don't have a choice in these things.

R82 Release Notes do state 8G (MINIMUM), and we all know what that means..

[Vincent_Bacher]

I understand that sometimes there is no other option.
Of course, it's different in production.
When it comes to management, I would always go big rather than small, if possible. That's why our two main MDS have 502 GB of RAM.

[Don_Paterson]

Yes!

CPM (java app) and postgres are resource hungry and lots of CPU and RAM keep them happy it seems.

High speed storage is good too 😉

[the_rock]

Btw Don, just did quick comparison, in R81.20 swap is about 100 MBs and R82 180 MBs, so not a big difference. They both use 12 GB of ram.

[Don_Paterson]

Cool, so there is a difference though, and you are between the 8G minimum RAM and 16G preferred/popular minimum (lets just call it that for now 🙂 )

[the_rock]

Well, thats true, BUT...there is always but word somewhere : - )

The thing is, in my case, difference seems pretty small considering R82 is a major version..in your case, thats a huge discrepancy.

[Don_Paterson]

Yes, there is, but 😉

In all seriousness. The 12G maybe could account for the small difference in your lab vs the larger difference in mine.

With 8G the new RHEL swap behaviour may be more aggressive than with 12G, and the result could be the much larger swap usage.

It's all just speculation. I am no black belt in Red Hat and wouldn't claim to be,

[the_rock]

haha : - )

Well, I get your point...Honestly, I would try with at least 12 GB and even possibly 16, if doable. I recall in the old days on windows Vista, when they used to recommend minimum amount of RAM and then you realize the next day, yes, that will NOT work lol

Maybe not the best comparison, but you get the point.

[Don_Paterson]

He said Windows Vista!! Nooooo... 😉

We're stuck with 8G for now but may get more.

Appreciate your input. 🙏

[the_rock]

I feel like some days as soon as I hear Windows anything, Im running away haha

[Vincent_Bacher]

Speaking of Vista. A long time ago, there was a very nasty meme, no idea if it would still be politically correct. Hiroshima 45, Chernobyl 86, Windows 98.

[the_rock]

geesh, crazy...

[Don_Paterson]

R81.20

R82

[the_rock]

Hey Don,

I cant tell, but is this top command or something else?

[Vincent_Bacher]

htop

[the_rock]

Never knew of that command, learn something new every day...let me test it shortly.

[Don_Paterson]

As always, hit h to see the help/keystroke option in there.

But what is also cool is that you can click on the column headers (e.g. MEM%) and sort by column.

CPU% is highlighted be default.

[Don_Paterson]

UPDATE:

I upgraded the R81.20 to R82 and the swap usage was higher straight away. At 320M but will monitor to see where it goes to.

[the_rock]

I would definitely try 12 GB of ram if you can.

[the_rock]

Just to try this in my own lab, I may change ram on R82 mgmt server, reboot and see what happens.

[Don_Paterson]

Great. Thanks.

After the upgrade I saw that jump to 320MB swap but then I rebooted and it stayed at 0 for the first hour, where the other lab goes up steadily.

This morning I went into CPUSE and started a package export (to save downloading R82 upgrade package for the secondary server) and the swap jumped up.

It stopped at around 430MB but it felt the need to swap and there was a burst of so (swap out).