1

Enable memory interleaving (disable NUMA) on 4-way Dell PowerEdge R820

view story
linux-howto

http://serverfault.com – We have recently received a new Dell PowerEdge R820 system. It comes with 4 CPU sockets and 512GB of memory. By default it operates in NUMA mode (memory interleaving is disabled) and practically each CPU socket sees 128GB of fast access memory, and 384GB of slow, remote access memory. As we run a java application with a large heap close to 512GB, and java has no serious support for NUMA, we would like instead to operate in (4-way) memory interleaving. But in the bios of the R820 this option is not available (you can see it but it is greyed). Is there a solution to enable memory interleaving (HowTos)

xskl's picture
Submitted by xskl on 10/11/2012

Stories similar to Enable memory interleaving (disable NUMA) on 4-way Dell PowerEdge R820

Is there a way to check whether node interleaving is enabled from within Windows on a Dell R710? omreport chassis biossetup doesn't appear to print any NUMA-related settings on the server I'm looking at.

We have a java application where towards the last part of the codes we wrote these lines

Runtime runtime = Runtime.getRuntime(); long memory = runtime.totalMemory() - runtime.freeMemory(); System.out.println("\n\nUsed memory is bytes: " + memory);

What we notice is that over time the top shows increase in the memory(%) column for the application but the java memory(from runtime variables) valu

I want to build a very parallel (at least 24 cores) AMD opteron (bulldozer) system.

I'm looking at some motherboards and I can't figure out if they're UMA or NUMA (uniform memory access vs non-uniform memory access.)

Are most multi-socket motherboards one or the other architecture?

Here are two I was looking at:

Quad socket Tyan: http://www.newegg.com/Product/Product.aspx?Item=N82E16813151219

numactl --membind 19 weeks 2 days ago

When using numactl in Linux with the --membind option, let's say I do the following:

numactl --membind=0,1,2 ./prog

Will the memory for ./prog be allocated on all of NUMA nodes 0, 1, and 2? Or will the memory be allocated only on NUMA nodes 1 and 2 if NUMA node 0's memory is not enough? Thanks.

We recently needed to add more RAM to our vCenter Server (Dell PowerEdge 860 server). We checked from the Kingston memory search what kind of memory the server accepted.

Then we found a listing of 4x 2GB DDR2-667 PC2-5300 ECC Registered Memory HP ProLiant DL180 G5 240-PIN on eBay. This looked like it would work.

I am having issues with a server running out of physical memory and I'm having troubles discerning whether it is from my application's Java process or something else on the server.

I have the hs_err_pid of a long running java process that has received a SIGSEGV. In C.6 System Section I can get a rough approximation of the free memory on the system from (MemFree + Buffers + Cached) / MemTotal, but I would like to find out exactly how much memory this process was addressing at the time of the crash, is this captured in this file?

I play with a java benchmark (SPECJbb) but observed some degrade with large pages (2M per page) enabled. Our server is NUMA arch based, has 80 cores (160 logical cpu when HT enabled) and 260G memory.

The benchmark consumes about 200G memory, so I allocated 200G large in kernel large page pool.But the performance is 6%~8% decline.

I have an LG -E405 phone running Android 2.3.6.

As can be seen here, the phone has 1GB internal memory and 384Mb RAM.

When I look at default App manager, this is what I get,

RAM - 231Mb System Memory - 157 Mb. Internal Memory - 378Mb. SD Card - 2Gb.

Also, all the applications I install, by default take up space from System memory.

Many tutorials explain about moving apps from internal memor