[SOLVED] ZFS Memory Hog ?

[KatyComputer]

I am migrating from a decent server to an awesome server, the decent server has 32GB memory, the guests are shut down, all I am doing is copying /var/lib/vz/dump to a local storage device. The system is burning 18GB. How do I know if I had something seriously misconfigured?







root@stl-eno1:~# lshw -short
H/W path Device Class Description
===================================================
system Super Server (To be filled by O.E.M.)
/0 bus X11SSH-F
/0/0 memory 64KiB BIOS
/0/2a memory 128KiB L1 cache
/0/2b memory 128KiB L1 cache
/0/2c memory 1MiB L2 cache
/0/2d memory 8MiB L3 cache
/0/2e processor Intel(R) Xeon(R) CPU E3-1230 v6 @ 3.50GHz
/0/2f memory 32GiB System Memory
/0/2f/0 memory [empty]
/0/2f/1 memory 16GiB DIMM DDR4 Synchronous 2400 MHz (0.4 ns)
/0/2f/2 memory [empty]
/0/2f/3 memory 16GiB DIMM DDR4 Synchronous 2400 MHz (0.4 ns)
/0/100 bridge Intel Corporation
/0/100/1 bridge Skylake PCIe Controller (x16)
/0/100/1/0 storage Samsung Electronics Co Ltd



The new box is a ProLiant DL325 Gen10 (P16696-B21) with 64GB RAM, 8 SAS SSD drives in a RAID 6 array powered by HP's P408i-a SR Gen10 RAID controller. So far, so good.

 

[aaron]

ZFS will by default use up to 50% of the RAM (if free) for the cache (ARC) to satisfy read requests from it. You can use the arcstat CLI tool to get detailed information. arcstat -v shows a list of all possible columns and their definition.

You can also set up some external performance monitoring and keep and eye on the arc size and the hit rate. The higher the hit rate is, or the lower the miss% is, the more read request could be delivered from the way faster ARC and access to the slow disks wasn't necessary.

You can limit the size of the ARC (search the forum/internet) but if possible I would try to get a hit rate close to 100% (miss% 0) during normal operation. If a guest is installing updates and such, where it will access data on disk that it usually doesn't touch, the hit rate will go down temporarily.