Capacity (GB or TB)
Performance(MBytes persec)
Throughput (I/O persec)
Latency (Milleseconds)
7200 RPM SATA :80 i/o ps
10K RPM SATA/SAS/FC :1200 i/o ps
15k RPM SAS/Fiberchannel :180 i/o ps
SSD (Solid State Device) MLC :1000-2000 i/o ps
EFD (Entp Flash Drive) SLC Single cell Tech :6000-30,000 i/ops
VMFS filesystem in SAN,iSCSI,FC ;NFS filesystem in NAS.
Blocked Based Storage:-SAN Fiber Channel (1- 20 GB speed) or iSCSI (100MB to 10 GB speed)
File Based Storage:-NFS (100 to 10 GB speed)
VMFS is used on local SCSI, FC, or iSCSI SAN (but not NFS)
SIOC(Storage I/O Control) is only implemented from 4.1
This will allows us to allocate share on per VM basis so that VM can access Storage I/O according
to the share allocated to them.
When Storage I/O Control is applied to Datastore cobfigured to multiple ESX hosts accessing a same stroage array Storage I/O Controller make sure that VM get it allocated i/o share.
Stroage Performance Factors:-
1)Number of VM on LUN
2)SAN/NAS configuration-Caching,RAID level, and number of spindles on a LUN
3)VMFS or RDMS RDMS offer no better performce than VMFS in thick mode
4) VMDK thin or thick
5)LUN queue depth(howbusy the LUN on a storage Array
6)Storage Protcol being used.
7) Storage I/O control
------------------------------------------------
Storage Performance :-
1)Average Disk Read Rate and Write Rate in KBps/disk (disk.read.average and disk.write.average)
2)Disk.devicelatency.average,disk.kernallatency.average,disk.queue.latency.average,disk.totallatency.average
per disk in millisecs
3)Device latency shouldn't exceed 25 ms,
Kernel and Queue latency shouldn't exceed 2 ms,
toal latency or command latency shouldn't exceed 27 ms
4)disk.commandsAborted.summation & disk.busRests.summation
Per disk number of commands aborted with Windows VMs,an aborted command will occur if the I/O command wasn't fulfilled in 60 seconds,by default
Any number > 0 indicates major storage I/O bottlenecks or physical storage issues
Esxtop:Davg(not exceed 25),Kavg,Gavg=dvag+Kavg(guest)total(not exceed 27),Qavg
u disk devices
F to change field
WWN :- 50:00:fn:jh:kj:th:kl First four 50:00 tell you the vendor name Qlogic,Emulex,CNA
CNA is only for FCOE.
VSAN is similar to VLAN .They Provide isloation between multiple logical SAN's exisitng
on a common physical Platform.
Zoning : To ensure that a LUN that is required to be visible to multiple hosts in a cluster.
Port based Zoning :
WWN base zoning :
Each VMFS volume has a metadata file which contains information about File locks.
SATP : SATP module handles path failover for a given storage array and determine the failover
type for LUN.
PSP :Handles the actual path used for every I/O .
How do we know VMFS datastore is overloaded ?
Q-length resxtop :U stroage QUED colum
Profile Driven Storage : Works with VASA (Vspshere Stroage API for storage awarness )
prinicple behind profile driven storage simples allow vsphere admins to build
VM storage profiles that describe the specific storage attribute that a VM requires.
Vspshere 5 has a software ISCSI intiator.
Once the LUN is presented you have to go and create a VMFS datastore.
Expanding VMFS Datastore :
1)You can dynamically expand the VMFS datastore if you have space avalible in LUN.
2)You can add extent if you do not have space in the LUN and then increase the VMFS volume
3)Each VMFS has a UNIQUE UUID
4)Choose keep exsisting signature if you want to mount the datastore copy without writing
a new signature only possible if the orginal is deleted or unmounted becoz of similar UUID
5)Choose Assign a new signature if you want to write a new signature onto the VMFS datastore
keep in mind this operation is irrversiable .If you have resignaured the datstore containing VM's
you will likely need to reregister those VM's in the Vcenter Server since path is changed.
No-root Squash option will give full explicit read/write acess.
Virtualize your dreams by Virtulizing your Applications,Servers,Networks and Desktops. Welcome to the Era of Cloud Computing .Go green in all the Datacenters.
Monday, August 29, 2011
Monday, August 8, 2011
Memory Management in VM
Memory Over Commitment:-
Transperent Page sharing
Memoery Ballooning
Hypervisor Swapping
Memory Comperssion
----------------------------------------------------
On a physical server MMU is part of Mother board which allocates memoery to OS.
For Virtual machines just like the how Physical have dedicated Hardware MMU to manage memory on the OS.MMU is virtulized for VM using either software using Shadow page tables or Hardware Intel EPT(extended page tables) or AMD RVI (Rapid Virtulized Index).
When a VM starts it has no physical RAM allocated to it and RAM is allocated as needed and recliamed when not in use using Transperent Page sharing,Memoery Ballooning,Hypervisor Swapping and Memory Comperssion 4.1.
Hypervisor use Pmap for each VM.Pmap manages what the VM's has to the host Physical Memoery.
Shaow pages tables in Hypervisor handles what Guest VM and Host pysical and SPtables are used by the MMU
http://software.intel.com/en-us/articles/best-practices-for-paravirtualization-enhancements-from-intel-virtualization-technology-ept-and-vt-d/
Hardware Page Table Virtualization provides a hardware assist to memory virtualization, which includes the partitioning and allocation of physical memory among VMs. Memory virtualization causes VMs to see a contiguous address space, which is not actually contiguous within the underlying physical memory. The guest OS stores the mapping between virtual and physical memory addresses in page tables.
Because the guest OSs do not have native direct access to physical system memory, the VMM must perform another level of memory virtualization in order to accommodate multiple VMs simultaneously. That is, mapping must be performed within the VMM between the physical memory and the page tables in the guest OSs. In order to accelerate this additional layer of memory virtualization, both Intel and AMD have announced technologies to provide a hardware assist. Intel's is called Extended Page Tables (EPT), and AMD's is called Nested Page Tables (NPT). These two technologies are very similar at a conceptual level.
Transperent Page sharing:-
Vsphere reclaims memeory by removing redundant memoery pages with identical Content.This is similar to
Dedupe performed on Stroage volumes.Say if you have three VM with 2008 any identiacal meomory among the three is hashed so that memory can be reclaimed through transparent page sharing.
Memoery Ballooning:-
VMEMCTL is loaded into guest OS using VM's tools this puts presure on the Guest OS when hypervisor runs low on memoery .Guest OS invokes its memoery cleanup and reclimation process freeing up pages for the ballon so the physical memoery pages assigned to ballon are free for other VM's.
Hypervisor Swapping.:- more latency becoz host need to swap memeory to disk and free up the Page and memoery.
Memoery Comperssion:
Host memmory Pages swapped out are stored in a compressed format in the host memoery for faster access.
Transperent Page sharing
Memoery Ballooning
Hypervisor Swapping
Memory Comperssion
----------------------------------------------------
On a physical server MMU is part of Mother board which allocates memoery to OS.
For Virtual machines just like the how Physical have dedicated Hardware MMU to manage memory on the OS.MMU is virtulized for VM using either software using Shadow page tables or Hardware Intel EPT(extended page tables) or AMD RVI (Rapid Virtulized Index).
When a VM starts it has no physical RAM allocated to it and RAM is allocated as needed and recliamed when not in use using Transperent Page sharing,Memoery Ballooning,Hypervisor Swapping and Memory Comperssion 4.1.
Hypervisor use Pmap for each VM.Pmap manages what the VM's has to the host Physical Memoery.
Shaow pages tables in Hypervisor handles what Guest VM and Host pysical and SPtables are used by the MMU
http://software.intel.com/en-us/articles/best-practices-for-paravirtualization-enhancements-from-intel-virtualization-technology-ept-and-vt-d/
Hardware Page Table Virtualization provides a hardware assist to memory virtualization, which includes the partitioning and allocation of physical memory among VMs. Memory virtualization causes VMs to see a contiguous address space, which is not actually contiguous within the underlying physical memory. The guest OS stores the mapping between virtual and physical memory addresses in page tables.
Because the guest OSs do not have native direct access to physical system memory, the VMM must perform another level of memory virtualization in order to accommodate multiple VMs simultaneously. That is, mapping must be performed within the VMM between the physical memory and the page tables in the guest OSs. In order to accelerate this additional layer of memory virtualization, both Intel and AMD have announced technologies to provide a hardware assist. Intel's is called Extended Page Tables (EPT), and AMD's is called Nested Page Tables (NPT). These two technologies are very similar at a conceptual level.
Transperent Page sharing:-
Vsphere reclaims memeory by removing redundant memoery pages with identical Content.This is similar to
Dedupe performed on Stroage volumes.Say if you have three VM with 2008 any identiacal meomory among the three is hashed so that memory can be reclaimed through transparent page sharing.
Memoery Ballooning:-
VMEMCTL is loaded into guest OS using VM's tools this puts presure on the Guest OS when hypervisor runs low on memoery .Guest OS invokes its memoery cleanup and reclimation process freeing up pages for the ballon so the physical memoery pages assigned to ballon are free for other VM's.
Hypervisor Swapping.:- more latency becoz host need to swap memeory to disk and free up the Page and memoery.
Memoery Comperssion:
Host memmory Pages swapped out are stored in a compressed format in the host memoery for faster access.
Subscribe to:
Posts (Atom)