I have just observed that all my FreeBSD machines running on ESXi 5.5 slow to a crawl when there is heavy i/o on the host machine.
However linux machines are absolutely fine as well as the host itself.
I am also not just talking about a slow down of peak performance, but the machines I cannot even login to ssh due to massive packet loss.
Here is some data from 'esxtop'
Showing 4 VMs
Fang, Test, and HardenedBSD1 are all BSD machines, whilst Seed-Debian is linux.
Notice how Seed-Debian has 0.00 in SWCUR.
Also in the main esxtop screen VMWAIT was about 500% for each BSD hypervisor but 0% for debian.
Apparently a high SWCUR figure means high swap activity, but ironically 2 of the BSD machines have all their ram reserved meaning they should never ever be swapped, debian has zero reserved. So something odd is going on here, and I am wondering if anyone has seen this before or has any ideas.
According to the performance graphs the BSD machines have no swapping as balloon reports zero, and I understand balloon to be a measure of ram used in the swapfile. However all the VMs report 0 for this yet esxtop reports 1069 MB of swap and 654 as cur.
I am guessing GRANT is physical ram assigned, so it would seem ESXi is favouring linux machines and ignoring the reservation settings completely, but I got no idea why.
However linux machines are absolutely fine as well as the host itself.
I am also not just talking about a slow down of peak performance, but the machines I cannot even login to ssh due to massive packet loss.
Here is some data from 'esxtop'
Showing 4 VMs
Fang, Test, and HardenedBSD1 are all BSD machines, whilst Seed-Debian is linux.
Notice how Seed-Debian has 0.00 in SWCUR.
Code:
4:11:08pm up 404 days 5:23, 503 worlds, 4 VMs, 14 vCPUs; MEM overcommit avg: 0.00, 0.00, 0.00
PMEM /MB: 16291 total: 1238 vmk,8633 other, 6419 free
VMKMEM/MB: 16212 managed: 651 minfree, 7761 rsvd, 8450 ursvd, high state
PSHARE/MB: 760 shared, 55 common: 705 saving
SWAP /MB: 1091 curr, 678 rclmtgt: 0.00 r/s, 0.00 w/s
ZIP /MB: 67 zipped, 43 saved
MEMCTL/MB: 0 curr, 0 target, 2571 max
GID NAME MEMSZ GRANT SZTGT TCHD TCHD_W SWCUR SWTGT SWR/s SWW/s LLSWR/s LLSWW/s OVHDUW
1732885 Fang 4096.00 3605.96 3951.73 573.44 573.44 492.84 138.11 0.00 0.00 0.00 0.00 19.37
12009 Seed-Debian 4096.00 4096.00 3855.95 122.88 81.92 0.00 0.00 0.00 0.00 0.00 0.00 6.64
187709107 Test 1024.00 622.61 447.04 61.44 40.96 254.29 248.09 0.00 0.00 0.00 0.00 7.17
39964045 HardenedBSD1 1024.00 665.00 705.37 61.44 30.72 344.06 292.58 0.00 0.00 0.00 0.00 7.17
Also in the main esxtop screen VMWAIT was about 500% for each BSD hypervisor but 0% for debian.
Apparently a high SWCUR figure means high swap activity, but ironically 2 of the BSD machines have all their ram reserved meaning they should never ever be swapped, debian has zero reserved. So something odd is going on here, and I am wondering if anyone has seen this before or has any ideas.
According to the performance graphs the BSD machines have no swapping as balloon reports zero, and I understand balloon to be a measure of ram used in the swapfile. However all the VMs report 0 for this yet esxtop reports 1069 MB of swap and 654 as cur.
I am guessing GRANT is physical ram assigned, so it would seem ESXi is favouring linux machines and ignoring the reservation settings completely, but I got no idea why.