This document discusses using esxtop and resxtop tools to troubleshoot performance issues on VMware ESXi hosts. It provides 10 key things to know about esxtop counters and how they work. It then gives examples of using esxtop to troubleshoot common problems like CPU contention, memory issues, network throughput problems, and disk I/O latency. It also lists some other diagnostic tools that can be used along with esxtop.

1. © 2010 VMware Inc. All rights reserved
Advanced performance troubleshooting using
esxtop/resxtop
Krishna Raj Raja
Staff Engineer, Performance Group

2. 2
Disclaimer
This session may contain product features that are
currently under development.
This session/overview of the new technology represents
no commitment from VMware to deliver these features in
any generally available product.
Features are subject to change, and must not be included in
contracts, purchase orders, or sales agreements of any kind.
Technical feasibility and market demand will affect final delivery.
Pricing and packaging for any new technologies or features
discussed or presented have not been determined.
“THESE FEATURES ARE REPRESENTATIVE OF FEATURE AREAS UNDER
DEVELOPMENT. FEATURE COMMITMENTS ARE SUBJECT TO CHANGE, AND
MUST NOT BE INCLUDED IN CONTRACTS, PURCHASE ORDERS, OR SALES
AGREEMENTS OF ANY KIND. TECHNICAL FEASIBILITY AND MARKET
DEMAND WILL AFFECT FINAL.”

3. 3
esxtop resources
esxtop manual:
http://www.vmware.com/pdf/vsphere4/r41/vsp_41_resource_mgmt.pdf
VMware Community documents:
http://communities.vmware.com/docs/DOC-9279 - ESX 4.0
http://communities.vmware.com/docs/DOC-11812 - ESX 4.1
esxtop for advanced users:
VMworld 2008 - http://vmworld.com/docs/DOC-2356
VMworld 2009 - http://vmworld.com/docs/DOC-3838

4. 4
Ten things that you need to know about
esxtop

5. 5
esxtop counters
1. esxtop does not create performance metrics
• esxtop derives performance metrics from raw counters exported in the
VMkernel System Info nodes (VSI nodes)
• esxtop can show new counters on older ESX system if the raw counters
are present in VMKernel

6. 6
esxtop counters
2. Counter values
• Many raw counters have static values that do no change with time – esxtop
displays them as it is
• Many counters increment monotonically, esxtop reports the delta for these
for the given refresh interval – for instance CMDS/sec, packets
transmitted/sec etc
• %USED and %RUN - CPU occupancy delta between successive
snapshots

7. 7
Refresh interval
3. Graphs will look different depending on the refresh interval
• Many counters values are dependent on refresh interval
• Larger refresh interval smoothens spikes and troughs
2 second refresh interval 10 second refresh interval

8. 8
esxtop counters
4. Counter normalization
• By default counters are shown for the group
• In group view counters values are cumulative
• In expanded view, counters are normalized per entity
Cumulative
stats
vcpu world
consumes CPU
Pressing ‘e’ key expands a group

9. 9
esxtop counters
5. %USED can exceed 100
• Turbo boost can increase the processor clock speed
• Asynchronous work can be happening on a different core on behalf of the
VM
VM on a NFS datastore running I/O intensive workload

10. 10
esxtop batch mode
6. Batch mode (-b)
• Produces windows perfmon compatible CSV file
• CSV file compatibility requires fixed number of columns on every row -
statistics of VMs/worlds instances that appear after starting the batch mode
are not collected because of this reason
• Only counters that are specified in the configuration file are collected, (-a)
option collects all counters
• Counters are named slightly differently

11. 11
esxtop batch mode – importing data into perfmon

12. 12
esxtop batch mode – viewing data in perfmon

13. 13
esxtop batch mode – trimming data
Trimming data
Saving data after trim

14. 14
esxplot
http://labs.vmware.com/flings/esxplot

15. 15
I/O Latencies
7. IO latencies
• IO latencies are measured per SCSI command so it is not affected by
refresh interval
• Reported latencies are average values for all the SCSI commands issued
within the refresh interval window
• Reported average latencies can be different on different screens (adapter,
LUN, VM), since each screen accounts for different group of I/Os

16. 16
resxtop – remote esxtop
8. You can use resxtop to connect to different ESX hosts
• Newer version of resxtop will connect to older ESX hosts
9. You don’t need root access to view esxtop counters
• resxtop can authenticate using vCenter credentials

17. 17
esxtop CPU usage
10. esxtop can consume non-trivial amount of CPU
• When you have very large inventory (VMs, LUNs, virtual disks, virtual NICs
etc)
• You can limit the amount of data collected by limiting the fields (columns)
and entities (rows), you can also reduce CPU consumption by locking
entities, (-l) option
CPU consumption on a host with 512 VMs
CPU consumption with esxtop -l
CPU usage when using resxtop

18. 18
Performance Troubleshooting Using
esxtop

19. 19
esxtop screens
Screens
• c: cpu (default)
• m: memory
• n: network
• d: disk adapter
• u: disk device (added in ESX 3.5)
• v: disk VM (added in ESX 3.5)
• i: Interrupts (new in ESX 4.0)
• p: power management (new in ESX
4.1)
VMkernel
CPU
Scheduler
Memory
Scheduler
Virtual
Switch
vSCSI
c, i, p m d, u, vn
VM VM VMVM

20. 20
Troubleshooting CPU Problems

21. 21
CPU Constrained
SMP VM
High CPU
utilization
Both the
virtual CPUs
CPU
constrained

22. 22
CPU Contention
4 CPUs,
all at
100%
3 SMP
VMs
VMs don’t
get to run
all the time
%ready
accumulates

23. 23
CPU Limit
Max
Limited
CPU
Limit AMAX = -1 : Unlimited

24. 24
Mis-configured SMP VM
vCPU 1 not
used by the
VM
Incorrect (UP) Kernel/HAL inside the
guest or the application inside the
guest is single threaded

25. 25
Power management – CPU frequency scaling
C states: C0 – busy, C1 – halted, C2 – deep halt
P states: P0 – Highest clock frequency, P11 – Lowest clock frequency

26. 26
VM Power Usage
Experimental feature, not enabled by default.
VMkernel advanced setting: Power.ChargeVMs

27. 27
CPU clock frequency scaling
%USED: CPU usage with reference to base clock frequency
%UTIL: CPU utilization with reference to current clock frequency
%RUN: CPU scheduled time
VM is running all
the time but uses
only 75% of the
clock frequency

28. 28
Hyperthreading
Two VMs running
on different cores
Two VMs sharing
the same core
%LAT_C counter
shows the time de-
scheduled due to
core sharing

29. 29
Timer interrupt rate
Linux Guests

30. 30
Timer interrupt rate
Windows Guests – Multimedia timer

31. 31
New metrics in CPU screen
%LAT_C : %time the VM was not scheduled due to CPU resource issue
%LAT_M : %time the VM was not scheduled due to memory resource issue
%DMD : Moving CPU utilization average in the last one minute
EMIN : Minimum CPU resources in MHZ that the VM is guaranteed to get
when there is CPU contention

32. 32
Troubleshooting Memory Problems

33. 33
esxtop memory screen (m)
Possible states:
high, soft, hard
and low
PMEM – Total Physical memory
VMKMEM - Memory managed by VMKernel
COSMEM - Memory used by Service Console

34. 34
Not able to power-on a new VM
Memory reservation
820 MB
reservation
requested
Overhead
memory
needs to be
reserved
4G memory
reservation

35. 35
Granted Memory
Granted Memory = Memory touched by the guest
Windows and FreeBSD Guests touches (zeroes) all its memory during boot
Linux Guests touches memory when it first uses it

36. 36
Ballooning versus Swapping
MCTL: N - Balloon
driver not active, tools
probably not installed
Memory
Hog
VMs
Swapped in
the past but
not actively
swapping
now
Swap target is
more for the VM
without the balloon
driver
VM with
Balloon
driver swaps
less

37. 37
Memory Compression Stats
COWH : Copy on Write Pages hints – amount of memory in MB that are
potentially shareable
CACHESZ: Compression Cache size
CACHEUSD: Compression Cache currently used
ZIP/s, UNZIP/s: Memory compression/decompression rate

38. 38
Wide NUMA - CPU
2 NUMA
nodes with
~6G each
NUMA home
node not assigned
6-vcpu VM –
cannot fit into
a NUMA node
size of 4
CPUs
4G, can fit into
a single node

39. 39
NUMA affinity not set
NUMA machine
with 2 nodes
CPU affinity set to
wrong NUMA node
All the memory in
remote node
NHN: NUMA Home Node
NLMEM: Memory in local node
NRMEM: Memory in remote node

40. 40
Wide NUMA - Memory
2 NUMA
nodes with
~6G each
NUMA home
node not
assigned
VM cannot be
fit into a single
NUMA node

41. 41
Troubleshooting Network Problems

42. 42
vSwitch active uplink
TEAM-PNIC : The uplink that the virtual switch port is currently using

43. 43
Dropped packets at vSwitch
Packet drops usually happens when the traffic has
no flow control (UDP/Multicast/Broadcast packets)

44. 44
Multicast/Broadcast stats
PKTTXMUL/s – Multicast packets transmitted per second
PKTRXMUL/s – Multicast packets received per second
PKTTXBRD/s – Broadcast packets transmitted per second
PKTRXBRD/s – Broadcast packets received per second

45. 45
NFS stats
DAVG and KAVG is not available for network backed storage
GAVG – gives the end to end latency

46. 46
Troubleshooting Disk Problems

47. 47
Disk I/O latency
Host bus adapters (HBAs) -
includes SCSI, iSCSI, RAID,
and FC-HBA adapters
Latency stats from the
Device, Kernel and the
Guest
DAVG/cmd - Average latency (ms) from the Device (LUN)
KAVG/cmd - Average latency (ms) in the VMKernel
GAVG/cmd - Average latency (ms) in the Guest

48. 48
Problem with the disk subsystem
Bad
throughput
Good
throughput
Device Latency is
high - cache disabled
Low device
Latency

49. 49
Insufficient Queue depth
Non-zero
KAVG
Queuing at
the HBA

50. 50
FC bottleneck
‘v’ – VM view
‘u’ – device view
‘d’ – adapter view

51. 51
vStorage API for Array Integration (VAAI) stats
CLONE_RD, CLONE_WR: Number of Clone read/write requests
CLONE_F: Number of Failed clone operations
MBC_RD/s, MBC_WR/s – Clone read/write MBs/sec
ATS – Number of ATS commands
ATSF – Number of failed ATS commands
ZERO – Number of Zero requests
ZEROF – Number of failed zero requests
MBZERO/s – Megabytes Zeroed per second

52. 52
VAAI - virtual disk creation example
vStorage API for Array Integration (VAAI)

53. 53
SCSI reservation conflicts

54. 54
Other diagnostic tools

55. 55
Other diagnostic tools (1 of 2)
sched-stats and schedtrace
• vm-support -s/-S flag captures sched-stats
• vm-support -c flag captures scheduler trace – takes lot of disk space
memstats
• Provides detailed memory usage stats with resource pool hierarchy
ft-stats
• FT Virtual Machine stats
• Collected with vm-support –s/S flag

56. 56
Other diagnostic tools (2 of 2)
swatchStats
• Stopwatch stats for VMFS, SCSI events
vscsiStats
• Virtual machine SCSI disk I/O stats
• Provides histogram information for latency, IO size, inter-arrival time and
outstanding I/Os

57. 57
vscsiStats
Virtual scsi disk
handle ids -
unique across
virtual machines
World group
leader id
Virtual Machine
Name
# vscsiStats -l

58. 58
vscsiStats – latency histogram
# vscsiStats -p latency -w 118739 -i 8205
Latency in
microsecondsI/O
distribution
count

59. 59
vscsiStats – iolength histogram
# vscsiStats -p iolength -w 118739 -i 8205
I/O block size
Distribution
Count