Re: [pve-devel] [PATCH] auto balloning with mom algorithm implementation
Just uploaded a new implementation - should work better now. I have finally uploaded an auto-ballooning implementation for pvestatd. The algorithm uses the new 'shares' property to distribute RAM accordingly. Feel free to test :-) ___ pve-devel mailing list pve-devel@pve.proxmox.com http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
Re: [pve-devel] [PATCH] auto balloning with mom algorithm implementation
I have done some tests yesterday, with around 10vm, overcommiting the host memory a lot, (mixing windows/linux vm) and it's seem to works fine. I'll try to do more extensive test next week. - Mail original - De: Dietmar Maurer diet...@proxmox.com À: Alexandre Derumier aderum...@odiso.com, pve-devel@pve.proxmox.com Envoyé: Jeudi 20 Décembre 2012 14:02:38 Objet: RE: [pve-devel] [PATCH] auto balloning with mom algorithm implementation Hi Alexandre, I have finally uploaded an auto-ballooning implementation for pvestatd. The algorithm uses the new 'shares' property to distribute RAM accordingly. Feel free to test :-) - Dietmar ___ pve-devel mailing list pve-devel@pve.proxmox.com http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
Re: [pve-devel] [PATCH] auto balloning with mom algorithm implementation
The current algorithm has some bug - I will try to improve more. Not sure about it, But I have seen some vm more ballonning than other. (with same shares value). - Mail original - De: Dietmar Maurer diet...@proxmox.com À: Alexandre DERUMIER aderum...@odiso.com Cc: pve-devel@pve.proxmox.com Envoyé: Vendredi 21 Décembre 2012 09:21:13 Objet: RE: [pve-devel] [PATCH] auto balloning with mom algorithm implementation I have done some tests yesterday, with around 10vm, overcommiting the host memory a lot, (mixing windows/linux vm) and it's seem to works fine. I'll try to do more extensive test next week. The current algorithm has some bug - I will try to improve more. ___ pve-devel mailing list pve-devel@pve.proxmox.com http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
Re: [pve-devel] [PATCH] auto balloning with mom algorithm implementation
Ok,thanks, I'll try that this afternoon ! - Mail original - De: Dietmar Maurer diet...@proxmox.com À: Alexandre Derumier aderum...@odiso.com, pve-devel@pve.proxmox.com Envoyé: Jeudi 20 Décembre 2012 14:02:38 Objet: RE: [pve-devel] [PATCH] auto balloning with mom algorithm implementation Hi Alexandre, I have finally uploaded an auto-ballooning implementation for pvestatd. The algorithm uses the new 'shares' property to distribute RAM accordingly. Feel free to test :-) - Dietmar ___ pve-devel mailing list pve-devel@pve.proxmox.com http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
Re: [pve-devel] [PATCH] auto balloning with mom algorithm implementation
Just committed the ballooning stats patches. Ok, thanks. Also added a fix so that we can set the polling interval at VM startup. Great ! Any news to get all stats values in 1 qom get ? Juts uploaded a patch for that. ___ pve-devel mailing list pve-devel@pve.proxmox.com http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
Re: [pve-devel] [PATCH] auto balloning with mom algorithm implementation
Just committed the ballooning stats patches. Ok, thanks. Also added a fix so that we can set the polling interval at VM startup. Great ! Any news to get all stats values in 1 qom get ? - Mail original - De: Dietmar Maurer diet...@proxmox.com À: Alexandre Derumier aderum...@odiso.com, pve-devel@pve.proxmox.com Envoyé: Mercredi 12 Décembre 2012 14:27:07 Objet: RE: [pve-devel] [PATCH] auto balloning with mom algorithm implementation -Original Message- From: pve-devel-boun...@pve.proxmox.com [mailto:pve-devel- boun...@pve.proxmox.com] On Behalf Of Alexandre Derumier Sent: Montag, 10. Dezember 2012 14:06 To: pve-devel@pve.proxmox.com Subject: [pve-devel] [PATCH] auto balloning with mom algorithm implementation for test! require ballooning stats patchs on top qemu-kvm Just committed the ballooning stats patches. Also added a fix so that we can set the polling interval at VM startup. https://git.proxmox.com/?p=pve-qemu-kvm.git;a=commit;h=904a90ccf761b43af8b6cae21be3a643cb228ae3 ___ pve-devel mailing list pve-devel@pve.proxmox.com http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
[pve-devel] [PATCH] auto balloning with mom algorithm implementation
for test!
require ballooning stats patchs on top qemu-kvm
Signed-off-by: Alexandre Derumier aderum...@odiso.com
---
PVE/QemuServer.pm | 141 -
1 file changed, 140 insertions(+), 1 deletion(-)
diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm
index 81cc682..3df3f57 100644
--- a/PVE/QemuServer.pm
+++ b/PVE/QemuServer.pm
@@ -31,6 +31,25 @@ use Time::HiRes qw(gettimeofday);
my $cpuinfo = PVE::ProcFSTools::read_cpuinfo();
+# then we will consider the host to be under memory pressure
+my $pressure_threshold = 0.20;
+
+# If pressure threshold drops below this level, then the pressure
+# is critical and more aggressive ballooning will be employed.
+my $pressure_critical = 0.05;
+
+# This is the minimum percentage of free memory that an unconstrained
+# guest would like to maintain
+my $min_guest_free_percent = 0.20;
+
+# Don't change a guest's memory by more than this percent of total memory
+my $max_balloon_change_percent = 0.5;
+
+# Only ballooning operations that change the balloon by this percentage
+# of current guest memory should be undertaken to avoid overhead
+my $min_balloon_change_percent = 0.0025;
+
+
# Note about locking: we use flock on the config file protect
# against concurent actions.
# Aditionaly, we have a 'lock' setting in the config file. This
@@ -1966,6 +1985,9 @@ sub vmstatus {
$d-{diskread} = 0;
$d-{diskwrite} = 0;
+ $d-{freemem} = undef;
+ $d-{balloon} = $conf-{balloon} ?($conf-{balloon} * 1024 *1024) :
undef;
+
$res-{$vmid} = $d;
}
@@ -2042,10 +2064,28 @@ sub vmstatus {
$res-{$vmid}-{diskwrite} = $totalwrbytes;
};
+my $freememcb = sub {
+ my ($vmid, $resp) = @_;
+ my $value = $resp-{'return'} || 0;
+ $res-{$vmid}-{freemem} = $value;
+};
+
+my $totalmemcb = sub {
+ my ($vmid, $resp) = @_;
+ my $value = $resp-{'return'} || 0;
+ $res-{$vmid}-{mem} = $value;
+};
+
my $statuscb = sub {
my ($vmid, $resp) = @_;
$qmpclient-queue_cmd($vmid, $blockstatscb, 'query-blockstats');
+ if( $res-{$vmid}-{balloon}){
+ $qmpclient-queue_cmd($vmid, undef, 'qom-set',path=
machine/peripheral/balloon0, property = stats-polling-interval, value =
10);
+ $qmpclient-queue_cmd($vmid, $freememcb, 'qom-get',path=
machine/peripheral/balloon0, property = stat-free-memory);
+ $qmpclient-queue_cmd($vmid, $totalmemcb, 'qom-get',path=
machine/peripheral/balloon0, property = stat-total-memory);
+ }
+
my $status = 'unknown';
if (!defined($status = $resp-{'return'}-{status})) {
warn unable to get VM status\n;
@@ -2068,9 +2108,109 @@ sub vmstatus {
$res-{$vmid}-{qmpstatus} = $res-{$vmid}-{status} if
!$res-{$vmid}-{qmpstatus};
}
+#auto-balloning
+my $hostmeminfo = PVE::ProcFSTools::read_meminfo();
+my $hostfreemem = $hostmeminfo-{memtotal} - $hostmeminfo-{memused};
+my $host_free_percent = ($hostfreemem / $hostmeminfo-{memtotal});
+
+warn host free:.$hostfreemem. total:.$hostmeminfo-{memtotal}.\n;
+
+foreach my $vmid (keys %$list) {
+ if($res-{$vmid}-{pid} $res-{$vmid}-{balloon}
$res-{$vmid}-{freemem}){
+ warn vm $vmid: mem:.$res-{$vmid}-{mem}.
maxmem:.$res-{$vmid}-{maxmem} . freemem:.$res-{$vmid}-{freemem}.
min_mem: .$res-{$vmid}-{balloon}.\n;
+if ($host_free_percent $pressure_threshold){
+balloon_shrink_guest($vmid, $host_free_percent,
$res-{$vmid}-{mem}, $res-{$vmid}-{maxmem}, $res-{$vmid}-{freemem},
$res-{$vmid}-{balloon});
+}else{
+balloon_grow_guest($vmid, $host_free_percent,
$res-{$vmid}-{mem}, $res-{$vmid}-{maxmem}, $res-{$vmid}-{freemem},
$res-{$vmid}-{balloon});
+}
+}
+
+}
+
return $res;
}
+sub balloon_shrink_guest {
+my ($vmid, $host_free_percent, $balloon_cur, $balloon_max, $freemem,
$min_mem) = @_;
+
+my $guest_free_percent = undef;
+ # Determine the degree of host memory pressure
+if ($host_free_percent = $pressure_critical){
+# Pressure is critical:
+# Force guest to swap by making free memory negative
+$guest_free_percent = (-0.05 + $host_free_percent);
+}else{
+# Normal pressure situation
+# Scale the guest free memory back according to host pressure
+$guest_free_percent = ($min_guest_free_percent * ($host_free_percent /
$pressure_threshold));
+}
+
+# Given current conditions, determine the ideal guest memory size
+# $guest_used_mem = $guest.StatAvg balloon_cur) - (guest.StatAvg
mem_unused);
+
+my $guest_used_mem = $balloon_cur - $freemem; # do we need average ?
+
+my $balloon_min = $guest_used_mem + ($guest_free_percent * $balloon_cur);
+
+$balloon_min = $min_mem if $balloon_min $min_mem;
+
+# But do not change it too fast
+my $balloon_size = $balloon_cur * (1 -
