Skip to contentSkip to conversation

xend: fix best NUMA node allocation

PATCH In review Andre Przywara 3 participants 3 messages Apr 15, 2010 – Jun 1, 2010

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xend: fix best NUMA node allocation

Hi,

since we moved several NUMA info fields from physinfo into separate functions/structures, we must adapt the node picking algorithm, too. Currently xm create complains about undefined hash values.
The patch uses the new Python xc binding to get the information and create a reverse mapping for node_to_cpu, since we now only have a cpu_to_node field.

Signature · 3 lines
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany
Tel: +49 351 448-3567-12
tools/python/xen/xend/XendDomainInfo.py +17 -12
diff -r 2c2591185f8c tools/python/xen/xend/XendDomainInfo.py
@@ -2711,7 +2711,7 @@
         else:
             def find_relaxed_node(node_list):
                 import sys
-                nr_nodes = info['max_node_id']+1
+                nr_nodes = info['max_node_index'] + 1
                 if node_list is None:
                     node_list = range(0, nr_nodes)
                 nodeload = [0]
@@ -2724,35 +2724,40 @@
                         if sxp.child_value(vcpu, 'online') == 0: continue
                         cpumap = list(sxp.child_value(vcpu,'cpumap'))
                         for i in range(0, nr_nodes):
-                            node_cpumask = info['node_to_cpu'][i]
+                            node_cpumask = node_to_cpu[i]
                             for j in node_cpumask:
                                 if j in cpumap:
                                     nodeload[i] += 1
                                     break
                 for i in range(0, nr_nodes):
-                    if len(info['node_to_cpu'][i]) == 0:
+                    if len(node_to_cpu[i]) == 0:
                         nodeload[i] += 8
                     else:
-                        nodeload[i] = int(nodeload[i] * 16 / len(info['node_to_cpu'][i]))
+                        nodeload[i] = int(nodeload[i] * 16 / len(node_to_cpu[i]))
                         if i not in node_list:
                             nodeload[i] += 8
                 return map(lambda x: x[0], sorted(enumerate(nodeload), key=lambda x:x[1]))
 
-            info = xc.physinfo()
-            if info['nr_nodes'] > 1:
-                node_memory_list = info['node_to_memory']
+            info = xc.numainfo()
+            if info['max_node_index'] > 0:
+                node_memory_list = info['node_memfree']
+                node_to_cpu = []
+                for i in range(0, info['max_node_index'] + 1):
+                    node_to_cpu.append([])
+                for cpu, node in enumerate(xc.topologyinfo()['cpu_to_node']):
+                    node_to_cpu[node].append(cpu)
                 needmem = self.image.getRequiredAvailableMemory(self.info['memory_dynamic_max']) / 1024
                 candidate_node_list = []
-                for i in range(0, info['max_node_id']+1):
-                    if node_memory_list[i] >= needmem and len(info['node_to_cpu'][i]) > 0:
+                for i in range(0, info['max_node_index'] + 1):
+                    if node_memory_list[i] >= needmem and len(node_to_cpu[i]) > 0:
                         candidate_node_list.append(i)
                 best_node = find_relaxed_node(candidate_node_list)[0]
-                cpumask = info['node_to_cpu'][best_node]
-                best_nodes = find_relaxed_node(filter(lambda x: x != best_node, range(0,info['max_node_id']+1)))
+                cpumask = node_to_cpu[best_node]
+                best_nodes = find_relaxed_node(filter(lambda x: x != best_node, range(0,info['max_node_index']+1)))
                 for node_idx in best_nodes:
                     if len(cpumask) >= self.info['VCPUs_max']:
                         break
-                    cpumask = cpumask + info['node_to_cpu'][node_idx]
+                    cpumask = cpumask + node_to_cpu[node_idx]
                     log.debug("allocating additional NUMA node %d", node_idx)
                 for v in range(0, self.info['VCPUs_max']):
                     xc.vcpu_setaffinity(self.domid, v, cpumask)

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#1 lore raw

Even with this patch applied I'm still seeing the cpu_to_node key errors in the xen-unstable build:

[2010-05-28 02:14:42 3313] ERROR (xmlrpclib2:178) Internal error handling xend.domain.create
Traceback (most recent call last):
File "/usr/local/lib/python2.6/dist-packages/xen/util/xmlrpclib2.py",
line 131, in _marshaled_dispatch
response = self._dispatch(method, params)
File "/usr/lib/python2.6/SimpleXMLRPCServer.py", line 418, in _dispatch
return func(*params)
File "/usr/local/lib/python2.6/dist-packages/xen/xend/server/XMLRPCServer.py",
line 80, in domain_create
info = XendDomain.instance().domain_create(config)
File "/usr/local/lib/python2.6/dist-packages/xen/xend/XendDomain.py",
line 982, in domain_create
dominfo = XendDomainInfo.create(config)
File "/usr/local/lib/python2.6/dist-packages/xen/xend/XendDomainInfo.py",
line 106, in create

    vm.start()

File "/usr/local/lib/python2.6/dist-packages/xen/xend/XendDomainInfo.py",
line 470, in start
XendTask.log_progress(0, 30, self._constructDomain)
File "/usr/local/lib/python2.6/dist-packages/xen/xend/XendTask.py",
line 209, in log_progress
retval = func(*args, **kwds)
File "/usr/local/lib/python2.6/dist-packages/xen/xend/XendDomainInfo.py",
line 2530, in _constructDomain

    balloon.free(16*1024, self) # 16MB should be plenty
  File "/usr/local/lib/python2.6/dist-packages/xen/xend/balloon.py",
line 187, in free
    nodenum = xc.numainfo()['cpu_to_node'][cpu]

KeyError: 'cpu_to_node'

xen_major : 4
xen_minor : 1
xen_extra : -unstable
xen_caps : xen-3.0-x86_64 xen-3.0-x86_32p hvm-3.0-x86_32
hvm-3.0-x86_32p hvm-3.0-x86_64
xen_scheduler : credit
xen_pagesize : 4096
platform_params : virt_start=0xffff800000000000
xen_changeset : Sat May 22 06:36:41 2010 +0100 21446:93410e5e4ad8
xen_commandline : dummy=dummy console=com1 115200,8n1
dom0_mem=512M dom0_max_vcpus=1 dom0_vcpus_pin=true iommu=1,passthrough,no-intremap loglvl=all loglvl_guest=all loglevl=10 debug acpi=force apic=on apic_verbosity=verbose
cc_compiler : gcc version 4.3.3 (Ubuntu 4.3.3-5ubuntu4)
cc_compile_by : bedge

cc_compile_domain      :
cc_compile_date        : Tue May 25 14:51:02 PDT 2010
xend_config_format     : 4

-Bruce

Quoted 19 lines · Andre Przywara

Hi,

since we moved several NUMA info fields from physinfo into separate functions/structures, we must adapt the node picking algorithm, too. Currently xm create complains about undefined hash values.
The patch uses the new Python xc binding to get the information and create a reverse mapping for node_to_cpu, since we now only have a cpu_to_node field.

--
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany
Tel: +49 351 448-3567-12

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#2 lore raw
Quoted 2 lines · Bruce Edge

Even with this patch applied I'm still seeing the cpu_to_node key errors in the xen-unstable build:

This should be fixed by xen-unstable:21494.

Thanks,
Keir