vSwitch balancing UCS

Similar Questions

• Weird error in Console after reboot ESX 4.1

  I received the following in a box of ESX 4.1 recently restarted error message.

  To interact with the configuration file /etc/vmware/esx.conf error: could not try to lock
  file.  Another process has locked the file for more than 20 seconds.  The treatment
  now the lock is/usr/sbin/esxcfg-nas (2999).  This operation will end if it is
  Run again after the lock is released.

  Then, after seeing this twice on the console, I got the following error:

  option with the name of that cimenabled already exists

  If I look at this process, it is: "/ usr/sbin/esxcfg-nas - b»  I don't know what about the - b option.

  The server is offline.  If I kill this process esxcfg-nas, nothing happens, but I really didn't expect it would be.

  This isn't something I've met before.

  Someone saw something like that?

  Thank you

  Ted

  Looks like that to me, the host strives to back up storage on boot up and unable to reach the NFS server.  I'm curious, but why remove the vmnic0 would correct this problem.  It is possible that if the vSwitch balancing policy has not been defined route based on IP Hash and you had two uplinks that are connected to the same vSwitch, that communication between the host and the NFS server was interrupted during the mount operation.  By deleting the uplinks all traffic was forced to land a single connection.  Have you tried to restore the vmnic0 to configuration at this point?

• Hi ALL, did any attempt on the virtual computer NETWORK load balancing using HYPERV on UCS blades

  I try to configure the CASE server cluster by using the Unicast NLB on the virtual machine on different blades on the UCS, it works for awhile, then he abandoned packages.

  I heard that this screenplay of unicast is not supported in the UCS when she used END-host mode in the fabric interconnet...? any attempted before.

  Would it, I use the multicast mode is that something needs to be done on the FBI62020 or the LAN switch upstream. ??

  Header note I found on the implementation of UCS for mulitcast NLBL:

  Microsoft NLB can be deployed in 3 modes:

  Unicast

  Multicast

  IGMP multicast

  For series B UCS deployments, we have seen that the multicast and IGMP multicast work.

  IGMP multicast mode seems to be the more reliable deployment mode.

  To do this, the monitoring settings:

  All NLB Microsoft value "Multicast IGMP" nodes.  Important!  Check ths by logging into EACH node independently.  Do not rely on the MMC of NLB snap.

  An IGMP applicant must be present on the VLAN of NLB.  If PIM is enabled on the VIRTUAL LAN that is your interrogator.  UCS cannot function as applicant IGMP.  If an interrogator of functioning is not present, NLB IGMP mode will not work.

  You must have a static ARP entry on cheating it upstream pointing IP address Unicast NLB on the multicast MAC address NETWORK load balancing.  This need will set up, of course, on the VLAN of the NLB VIP. The key is that the routing for the NLB VLAN interface must use this ARP entry as a unicast IP ARP response may not contain a multicast mac address. (Violation of the RFC 1812)  Hosts on the NLB VLAN must also use the static entry.  You may have several entries ARP.  IOS can use a function of 'alias' of ARP. (Google it.)

  How Microsoft NLB works. -The truncated for brevity Mac addresses.

  TOPOLOGY OF NLB MS

  NETWORK VLAN 10 = subnet 10.1.1.0/24 IP load balancing

  VIP = 10.1.1.10 NETWORK LOAD BALANCING

  Arp entry static switch advanced IP 10.1.1.10 upstream to MAC 01

  NLB VIP (MAC 01, IP 10.1.1.10)

  NODE-A (AA, MAC IP:10.1.1.88)

  NŒUD-B (MAC BB, IP:10.1.1.99)

  Using the IGMP snooping and interrogator VLAN snooping table is filled with the mac NLB address and groups pointing to the appropriate L2 ports.

  MS NLB nodes will send the responses of IGMP queries.

  This snooping table could take 30 to 60 seconds to complete.

  Host on VLAN 200 (10.200.1.35) sends traffic to NETWORK VIP (10.1.1.10) load balancing

  It goes of course to VLAN 10 interface that uses the static ARP entry to resolve to address MAC 01 VIP NETWORK load balancing.

  Since it is a multicast frame destination it will be forward by the IGMP snooping table.

  The framework will arrive at ALL NLB nodes. (NŒUD-A & NŒUD-B)

  NLB nodes will use its load balancing algorithm to determine which node will manage the TCP session.

  Only one NLB node will respond to this host with TCP ACK to start the session.

  NOTES

  This works in a VMware with N1k, standard vSwtich and vDS environment. Where surveillance IGMP is not enabled, the framing for VIP MAC NETWORK load balancing will be flooded.

  NLB can only work with TCP-based services.

  As stated previously mapping an IP unicast to a multicast mac address is a violation implied by RFC 1812.

  TROUBLESHOOTING

  Make sure your interrogator is working. Just to clarify that this does not mean that it is actually at work.

  Wireshark lets check that IGMP queries are received by the NLB nodes.

  Make sure that the ARP response works as expected.  Once Wireshark again is your friend.

  Look at the paintings IGMP snooping. Validate the L2 ports appearing as expected.

  CSCtx27555 [Bug-preview for CSCtx27555] Unknown multicast with destination outside the range MAC 01:xx: are deleted. (6200 FI fixed in 2.0.2m)

  IGMP mode not affected.

  CSCtx27555    Unknown multicast with destination outside the range MAC 01:xx: are deleted.

  http://Tools.Cisco.com/support/BugToolKit/search/getBugDetails.do?method=fetchBugDetails&bugId=CSCtx27555

  fixed in 2.0(2m)

  Solution: Change the NLB mode of operation of "Multicast" to "multicast IGMP', which modifies balancing load NETWORK VIP MAC at 0100.5exx.xxx Beach, allows to transfer occur as expected.

  Q: and if I switch to switch mode, which means all of the profile and the settings on the servers are completely exhausted and I need to recreate them. ???

  A:Cisco Unified Computing System Ethernet switching Modes

  http://www.Cisco.com/en/us/solutions/collateral/ns340/ns517/ns224/ns944/whitepaper_c11-701962.html

  -There is no impact on the configuration, you have done service profiles.  they will continue to work as expected.  Mode selector has the FI behave more like a conventional switch.  Most notable is that Spanning tree will be activated and if you have several uplinks yew, tree covering weight will begin to block redundant paths.

  You need to review your topology and what impact tree covering weight.  Generally, we at the switch port upstream defined as "edge master", you want to delete this line.

  For pre-production and laboratory environment, PDI can help qualified with the planning, design and implementation partners.  Given to review the IDP site and open a case if you need more detailed assistance.

• Nexus 1000v, UCS, and Microsoft NETWORK load balancing

  Hi all

  I have a client that implements a new Exchange 2010 environment. They have an obligation to configure load balancing for Client Access servers. The environment consists of VMware vShpere running on top of Cisco UCS blades with the Nexus 1000v dvSwitch.

  Everything I've read so far indicates that I must do the following:

  1 configure MS in Multicast mode load balancing (by selecting the IGMP protocol option).

  2. create a static ARP entry for the address of virtual cluster on the router for the subnet of the server.

  3. (maybe) configure a static MAC table entry on the router for the subnet of the server.

  3. (maybe) to disable the IGMP snooping on the VLAN appropriate in the Nexus 1000v.

  My questions are:

  1. any person running successfully a similar configuration?

  2 are there missing steps in the list above, or I shouldn't do?

  3. If I am disabling the snooping IGMP on the Nexus 1000v should I also disable it on the fabric of UCS interconnections and router?

  Thanks a lot for your time,.

  Aaron

  Aaron,

  The steps above you are correct, you need steps 1-4 to operate correctly.  Normally people will create a VLAN separate to their interfaces NLB/subnet, to prevent floods mcast uncessisary frameworks within the network.

  To answer your questions

  (1) I saw multiple clients run this configuration

  (2) the steps you are correct

  (3) you can't toggle the on UCS IGMP snooping.  It is enabled by default and not a configurable option.  There is no need to change anything within the UCS regarding MS NLB with the above procedure.  FYI - the ability to disable/enable the snooping IGMP on UCS is scheduled for a next version 2.1.

  
  
  This is the correct method untill the time we have the option of configuring static multicast mac entries on
  
  the Nexus 1000v.  If this is a feature you'd like, please open a TAC case and request for bug CSCtb93725 to be linked to your SR.
  
  This will give more "push" to our develpment team to prioritize this request.
  Hopefully some other customers can share their experience.
  Regards,

  Robert

• Always load balancing algorithms "static mapping vSwitch port for Teddy adapter?

  I'm looking for a LBT for my move to 10 G ethernet and see this KB

  http://KB.VMware.com/selfservice/microsites/search.do?language=en_US & cmd = displayKC & externalId = 1022590

  It is said

  """

  In vSphere 4.0, there are three strategies:

  • Route based on the originating Virtual Port ID
  • Function hash IP route
  • Route in function interference source to the MAC

  These three policies provide a static mapping of vSwitch for Teddy adapter port.

  """

  I always thought that IP hash would allow a virtual machine to use more than one bear, so she sent to several IP addresses, am I wrong or the Ko?

  Tom

  Maybe it is not described as clearly as it could be. However, the traffic of a VM (IP address) to a specific target (IP address) is always sent over the same bear. That's what the article described as static.

  Since you are looking to balance the load. Have you already taken a glance to vSphere 4.1. It offers "focused on the burden of reunification." Maybe you want to read http://blogs.vmware.com/performance/2010/12/vmware-load-based-teaming-lbt-performance-.html for details and performance.

  André

• mode - the values defined in the balancing vSwitch batch file

  Hi gurus.

  I need to change to a long list of hosts of fashion for a new value for the instance "route based on the original port code" vSwitch load balancing.

  And as I'm a lazy VMware admin, I want to batch

  Someone has a tip for me?

  THX,

  Didier

  Hey Buddy,

  This is the command:

  vmware-vim-cmd /hostsvc/net/vswitch_setpolicy -nicteaming-policy=loadbalance_srcid vSwitchX

• vswitch\pswitch and NIC load balancing

  Hi all, I would like to pass a scenario by you all to see if I'm on the right track...

  We have 2 guests with DAS each and the vswitches to connect with 2

  1 GB card physical network of a pswitch (spine) of the spine, I have 3.

  2nd level going @ 100 MB / who joined forces 1 GB connection to the

  spine. With me so far?

  Now, I think that the network knows contension request that

  Sometimes we lag eg: switching shared printers printing

  Server, digging the network shares, etc...

  I have check the performance data on the host computer and I see not heavy

  bandwidth, the cpu, the disk activity load. I check the NETWORK card and it

  appears only 1 never get used even if I don't know they are associated

  correctly.

  I decieded to get another switch of the same spine model and plan

  to use the covering tree weight to connect backbones and with network cards

  Separate them each to the spine and then another tree covering weight

  in each office, idea that switching will be better

  spread the load across the two NIC to the vswitch...

  At present, there are only 2 NICs in each host, I have another 2

  cards Go transit and 2,100 mb NIC to use for the console. I plan to use 2 news

  NIC Go for interhost connectivity and create a virtual SAN for extra

  space on each DAS.

  I've attached diagrams of deouifinecomptent Rapids and the proposed crossing and also how I could team up on the vswitch...

  Thank you for your time!

  That's what I've got to meet the night - if you change your IP Hash the vmkernel will review each IP packet out and based on the IP address of the source and the destination the vmkernel wil choose the physical NIC connected to the virtual switch - so for an example, and I will keep it simple assumes that you don't have a simple VM with a single virtual NETWORK adapter connected to the virtual switch - now lets look at two examples possible

  1. The virtual machine is a web wean provideing conebt to multiple users - each IP address of the V will remain constant, but each user will have a different IP address for traffic coming out will be distributed in all of the cards of physoical connected to the virtual switch.

  2. The virtual machine is an application server to connect to a physical SQL Server - in this case the starting and destination IP addresses will not change if the traffic go on a physical network adapters

  Hope this cleared up-

  If you find this or any other answer useful please consider awarding points marking the answer correct or useful

• Are these viable designs for NIC teaming on UCS C-Series?

  Is this a viable design on ESXi 5.1 on UCS C240 with 2 Quad port nic cards?

  Option A) VMware consolidation of NIC with the load balancing of interfaces vmnic in an configuration active/active through alternate and redundant physical paths to the network.

  Option B) VMware consolidation of NIC with the load balancing of the interfaces in a configuration vmnic Active/Standy by alternate and redundant paths of material to the network.

  Option A:

  

  Option B:

  

  Thank you.

  She really comes down to what means active/active N ° and type of switches upstream.  For ESXi NIC teaming - Active/Active load balancing provided the opportunity for all network links be active for devices different comments.  Grouping can be configured in a few different methods.  The default value is by ID of virtual port where each guest computer is assigned to an active port and then also a spare port.  Traffic to this host would be only sent on a connection at a time.

  For example, suppose 2 Ethernet connections and 4 guests on the ESX host.  Link 1 for switch 1 would be active for 1 and 2 and 2 link to switch 2 backup for 1 and 2.  However 2 link to switch 2 would be active for 3 guests and 4 and 1 link 1 switch backup for 1 and 2 comments.

  The following provides details on the configuration of the card NETWORK is teaming up with VMWare:

  http://KB.VMware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalID=1004088

  There are also opportunities for LACP configuration in some situations, but there is a material special considerations on the side of the switch, but also on the host side.

  Keep in mind that the vSwitch does not blindly broadcast/multicast/unknown unicast on all ports.  He has a strict set of rules that prevents loopback.  It is not a traditional L2 redirector so loops are not taken into account in an active/active environment.

  In addition, this document explains VMWare Virtual Networking Concepts.

  http://www.VMware.com/files/PDF/virtual_networking_concepts.PDF

  Steve McQuerry

  UCS - Technical Marketing

• Cannot set policies Vswitch

  Hello

  I have added a field to VMM a Vcenter that is running on a chassis of UCS - B. Now, I get a few defects related to CDP/LLDP adjacency. I can modify the UCS and Vcenter. However, APIC, I can't submit any changes to the policy of Vswitch. When I go in the area of the WWW, I have the section political VSwitch and I can choose a different in the drop-down lists for port-channel, lldp, cdp etc but the remains of gray submit button, so where the changes are not saved.

  Is this a bug or is there something else?

  I am running v.2.1(1h) with a VMware VDS and Vcenter 6.0

  Thank you

  Roy

  Yes checked in the code later (not yet released) and is not a problem.  Chatty & confirmed with Dan he has this experience with your version of the code.

  Dan will ensure the bug ID (probably already resolved) for your records.  Workaround for now would be to recreate the domain VMM with the correct settings.

  See you soon,.

  Robert

• Generational UCS/VMWare

  Hello

  I'm looking for clarification on how the Palo VIC presents the vNIC, how say/control which card vNIC to which ports physical and how best to configure my environment for VMWare.

  Summary of material:

  Blades (full-width) B250 with 2 cards of Palo VIC

  5108 chassis with double FEX (double uplinks from each FEX)

  2 x 6120 controllers with 2 x 10 Gbit links rising each to a pair of Nexus s 5020

  Summary of the software:

  UCS 1.3.0 boot loader, the kernel/system 4.1 version (3) N2(1.2d)

  VMWare vSphere Enterprise Plus 4.1

  According to my interpretation, Cisco and VMWare recommend against bandwidth limiting the interfaces - but prefer to go with actions and ability to burst.

  Therefore, my desired construction is the following:

  for each of the 4 available in B250 blade 10gbit interfaces, I want to have 2 vNIC (1 for VM distributed vSwitch [6Gbit minimum actions, 802. 1 q trunks], 1 for iSCSI [4 Gbit minimum actions, access port/no 802. 1 q trunk]). This means that my Service profile must define 8 vNIC altogether.

  Q1: How can I make sure that I get 1 'data' and 1 'iscsi' vNIC per physical port?

  Q2: how to set a policy to set the minimum bandwidth on each vNIC?

  Q3: How do I control/make sure the binding order correct so that I can script my configuration of vSphere?

  Generation of VMWare will be vswitch distributed containing all the 'data' vNIC, allowing control of network IO with actions defined to manage the various allowances for groups of data ports, console and VMotion. The vNIC iSCSI will each receive a VMKernel port dedicated.

  Thank you very much in advance for any help.

  Our installation program is not the same, but the principles can be applied as we also CIV.  To make sure that we had the installation of ports properly, I put into service six vNIC in UCSM to each host.  I put in service the first vNIC of fabric and the second vNIC fabric B - do not check the box "Fabric Failover"...  Repeat for other network cards for four.  This will give you 4 vNIC by fabric.  Note MAC addresses and which tissue they are attached.

  Then, in case of VC and to configure your hosts, enter the network cards and line up the vmnic and MAC addresses so that you know how they trace.  When you configure your vSwitches, just make sure you add a vmnic from fabric and a fabric vmnic B.    Do this for the Service Console, vmkernel and VM switches.

  If you make this unique environment of VIC inas, it will ensure that your vswitches are attached to every tissue.  I guess you can do this same thing with two VICs, since each card may appear independently in UCSM.

  As for the bandwidth mgmt and others, check out http://bradhedlund.com/ its recent messages are on the networks of the UCS.

  Adam

• UCS 1.4 support for PVLAN

  Hi all

  Cisco UCS 1.4 Adviser supports PVLAN. But I see the following comment on PVLAN in point 1.4 of the UCS

  "UCS extends PVLAN support for virtualized (without vSwitch) deployments."

  "UCS release 1.4 (1) supports isolated PVLAN support for physical server paths or Palo NAC vNIC ports."

  Is that means PVLAN will not work for the virtual machine if VMs is connected to the UCS by Nexus1000v or vDS even if I use the card of PALO (M81KR)?

  Could someone can confirm that?

  Thank you very much!

  Hello

  Yes - PVLANs must be from end to end.

  You can extend the PVLAN set to 1000v and then to bare metal (attached to UCS) servers to a port promiscous * outside * UCS as a backup station or a L3 interface.

  Pls remember that promiscous ports are not supported in UCS. Only access isolated at the moment.

  With the M81KR adapter (Palo), you must create a diff for each interface isolated VLAN (as UCS cannot do isolated trunks yet).

  In you case you seem to have 1 isolated VLAN vNIC so 4 total giving to the 1000v running ESX.

  2 vNIC for VLAN regular (load sharing and redundancy)

  2 network cards for the VLAN isolated (load sharing and redundnacy).

  See attached a doc that talking abt doing which I think is what you're asking.

  Thank you

  -Matt

• Uplinks ESX on a blade of UCS

  After installing ESX on UCS Blade (adapter type is Palo, the number of network cards is 2) sometimes I see that the uplinks are vmnic2 and vmnic3 (output of esxcfg-NICS - l)

  Where are the output of esxcfg-vswitch - l shows the uplink as vmnic0, and it also generates a warning saying there is no such thing as vmnic0.

  This happens not all the time. Anyone know why this might be happening?

  I saw this happen a little when you do not have NICs as the first devices in the vNIC/vHBA Placement. By default, we put the HBAs first sometimes ESX doesn't like it and will the NPI as vmnic2 and 3. If you move the network above the HBA cards, they come as a vmnic0 and vmnic1. I have always to reorganize and put my NIC before my storage.

  Louis

• Cisco UCS C210M2-VCD2 network connectivity

  Hello

  I will install three servers of c210M2 and I am quite new to the world of the UCS. According to the design that I'm supposed to install

  4 x Vm (CUCM, CUC, presence, UCCX) primary or Publisher on NGC 01

  4 x Vm (CUCM, CUC, presence, UCCX) subscribed or failover on NGC 02

  2 x Vm (MeetingPlace) on NGC 03

  I have a few switches access X 3750 for network connectivity. My question is how to configure the NICs on UCS Server or the network. According to my knowledge each server has cards NIC 2 x 10 giga and 1 x NIC integrated management. I'm confused how many NICs each VM will have and how to set up all these VMs NIC on 2 x card physical NETWORK by UCS. Any suggestions please.

  Kind regards

  Asif

  Asif,

  Nexus 1000 would be useful here, but if not an option and that you use standard vSwitch then both NICS should serve rising in a unique vSwitch for redundancy.

  Here are some suggestions of QoS that might help you too.

  Rredundant of the physical LAN interfaces are recommended

  One or two pairs of cluster NETWORK adapter for the CPU VM traffic. A pair just on C210 due to the low load per VM

  A pair of NIC (or dedicated team) for specific traffic VMware (mgmt, vMotion, VMware HA, etc.)

  SIDENOTE: Pair of grouping NIC can be divided through the motherboard and the PCIe card permitting the UCS C-series, protects against the failure of the PCIe card

  Once we know exactly asked what vSwitch using as the last post, we are able to locate more, but this is based on the standard use of vSwitch

  I hope this helps.

• UCS Fiber Channel FI for San switch wiring & config

  My question is about the correct wiring & vSAN configuration by the UCS FI switch San (Cisco MDS).

  Currently my wiring is as follows:

  Fi1 port 25 - SW1 port 3

  Fi1 port 26 - SW2 port 3

  Fi2 port 25 - SW1 port 4

  Fi2 port 26 - SW2 4 port

  On the side of the switch San, our guys of storage configured vlan10 SW1 and SW2 as vlan20. For this reason, I was not able to have a vSAN by FI, I read is recommended, because the VLAN switch cross FIs. Now, when I created vHBAs I have 4 vHBAs per server, 2 by IFS by vlan if I want redundancy with a switch or a loss of FI.

  Question: Is this the correct configuration on the side of the SAN switch? Do I actually create 4 vHBAs per server? I do not have my incorrect wiring & I should just be linking FI1 to SW1 on two ports & FI2 to SW2?

  Help or advice would be appreciated!

  ~ Tara

  Hi Tara,

  There are two ways of wiring of your SAN infrastructure.  It boils down to a design decision. Traditionally the SAN admins prefer to keep their separate/isolated tissue.  In this design THAT MDS-1 only connect to FI - A and MDS-2 would connect only to the FI - B.

  Another option is to connect each MDS at each END.  This gives each FI SAN fabric access and allows to create double VSANS (that exists on each FI) rather than separate them.

  You don't need four vHBAs by UCS Server - some adapters support no more than two in any case.   What you do is create two vHBA by the Service profile, each host will have access both to VSAN 10 and 20.  What you can do is to shift your tissue of vHBA while 1/2 access your Service VSAN 10 via FI - profiles and the other half access VSAN 10 FI - B.  This will help to distribute the load accordingly.

  E.g.

  First 1/2 of SPs

  vHBA0 - VSAN10 - fabric-A

  vHBA1 - VSAN20 - fabric-B

  1/2 second of SPs

  vHBA0 - VSAN20 - fabric-A

  vHBA1 - VSAN10 - fabric-B

  As you can see a bit of a manual balancing can act.  The best scenario would be to isolate (re-cable) to your network SAN fabrics according to the traditional SAN design above.

  Kind regards

  Robert

• storage iSCSI with UCS

  Hi all

  Can I ask a question regarding the connection of iSCSI storage for use with UCS. Look at us with Nimble iSCSI based storage & want to understand best practice recommendations on how to connect it to UCS to get the best level of performance & reliability / resilience etc.

  Another issue is more closely how VMware deals with loss of connectivity on a path (where double connections are the installation program from the warehouse to the tissues), would he re - route traffic to the path running?

  Any suggestion would be appreciated.

  Kassim

  Hello Kassim,

  Currently the agile iSCSI storage is certified with UCS 2.0.3 firmware version.

  http://www.Cisco.com/en/us/docs/unified_computing/UCS/interoperability/matrix/r_hcl_B_rel2.03.PDF

  The following guide can serve as a reference.

  Virtualization solution Cisco with the agile storage reference Architecture

  http://www.Cisco.com/en/us/solutions/collateral/ns340/ns517/ns224/ns836/ns978/guide_c07-719522.PDF

  In above installation, ESXi software iSCSi multipath with PSP Round Robin algorithm is implemented to take care of the IO and failover with load balancing two paths.

  HTH

  Padma