UCS Introduction PDF

UCS Introduction Manageme nt 2009…CISCO try to find single interface to manage servers regardless of their numbers. Cables management was also an issue. Fabric Interconnect makes cable management easy and provide fast speed connections. If server replaced….MAC address changes which is b...

UCS Introduction Manageme nt 2009…CISCO try to find single interface to manage servers regardless of their numbers. Cables management was also an issue. Fabric Interconnect makes cable management easy and provide fast speed connections. If server replaced….MAC address changes which is bound to policies Similar thing with fiber channel storage connectivity WWNN(World- wide node name)---used in fiber security called zoning UCS solve this problem with services profiles and policies created in software. Application point of view no change even hardware changes UCS Design what servers? Small Form Factors No cable sat back  plug and go 8 servers in single chassis with 6 RU Power modular at bottom—efficient power For compact design UCS Design what servers? Small Form Factors No cable sat back  plug and go Independent put in rack and cable to network can put PCI card on back to enhance capabilities Up-to 24 hard drives support For flexibility UCS Design what servers? Up-to 56 Hard Drives with 10TB per drive 4 RU and PCI cards from back possible For data backup  Fabric Interconnects? Servers connection to network  problem FI between servers and network S and C series can be connected to network directly, but B-series must be connected to FI UCSM lives on FI to manage all servers UCS Manager Manage all servers and changes from single point Comes free with system. Preinstalled on FI Accessible via VIP VIP Linked to both FI Can be accessed via any browserHTML5 based UCS-Out of Band access Now UCSM in cloud Internight How it looks like? Service profiles Repair ? Server ? Physical –mborard, cpu, mem,hdd. NICs, HBAs toc connect Whats not Hardware  MAC addresses, WWNN, UUID-motherboard, Policies (Bios setting), firmware, RAID policies etc, Boot policies Service profile identity+ policies Server exists separately from hardware. Service profile can be applied/Copied to new server Building Blocks of UCS IOMs (Fabric Extenders)expanding networking 2208 IOMs (Fabric Extenders)expanding networking via proper cable + management from FI’s UCSM 2208 Fabric Interconnects Handles networking traffic Nexus-switches—but can act as EHV mode Can not connect together Between LAN –Upstream and Servers-Downstream Runs in two modes Switch mode –not suitable for data center End host virtualization (EHV) mode Virtualizing to network that they are end hosts No STP protocols and layer-2/3 switching protocols MAC learning just downstream …not upstream Traffic from server to network---pinning applied (port channel or pin each server traffic to specific uplink port) Traffic from server is sent to specific upstream link if its is not destined for other server in downstream (bounded with server) Pinning Upstream networking FI connect to data center switches (Nexus) to connect to LAN Normally Switches can be connected with different connections between but Fis are also not like switches. Connections between two FIs not possible They connects to nexus switch which in turn can be connected together All connection coming to FI can be bundled together (Virtual port channel) but we can bundle two-two together if 4 lines are connected to FI (Two VPC) but with pinning no need of this LACP is negotiation protocol for port channels in UCS VLANs configuration n UCS VLANs enabled on all links from FI by default. To change it, create VLAN groups and assigned to different link (Layer 2 disjoint networks) Set one link on FI as Designated receiver per VLAN to receive broadcast (avoid sender to receive it back and server to receive two copies) Quality of service Prioritizing traffic like application’s data storage > backup/updates traffic. How switch identifies traffic via tagging Ethernet header has a field COS(Class of service) only on trunk links (In UCS everything is trunk) Tags 0---5 (6,7 for control traffic) 0best effort (backup) , 3 storge traffic, 2 live VM migration In UCS, COS can be mapped like 50%COS 2, 50%COS 0 on FI/Switch Quality of service FI has FCoE (FC protocol inside Ethernet)COS 3 by default andis 50% MTU—Max size of frame to hit interface on ethernet In UCS MTU>=9000 (Jumbo frames), header negligible in this case, more efficient utilization CISCO vNIC Hypervisors like multiple disparate connections PROBLEM: For 12 servers running hypervisors, each requires 10G connection=>12 10G ports too expensive Solution: vNIC Two 10 G connection can be used as many 10G virtually Two VMs on same host can not communicates with each other via normal physical switch, so vSwitch is needed. Physical connection will be terminated at Vswitch Multiple vSwitches on host due to separated traffic To have multiple virtual switches necessitates multiple physical connections  CISCO vNIC For UCS Servers, CISCO creates VICs CISCO vNIC  CISCO vNIC Virtual ethernet port can work as switchport to connect vlans to act as trunk etc UCSM vNIC connected to Veth Interface Think it as switchport configuration It is at vNIC level in UCS because vNIC connected to veth interface at FI Resource pools 256 vNICs by CISCO What about so many MAC addresses? Resource pool of MAC addresses can be created and those assigned automatically to vNICs Lets 160 servers, each 10 vNICs , nightmare to allocate manually 1600 MAC 6 octets , first three OUIfor UCS it is 00:25:B5, Last three 00:00:00 to 00:00:63 (63=99 in dec) 100 MAC address pool Can create two different pools as well like 00:25:B5: AA:00:00  00:25:B5: AA:00:63 00:25:B5: BB:00:00  00:25:B5: BB:00:63 Service profile We create service profile with say, 10 vNICs and assign each to specific MAC pool (FI-A and FI-B) Service profile can be created/copied and vNICs will be attached with the same pools Like MAC addresses, pools for other identity points can be created like IP addresses. Storage networking protocols Server  Storage Array (low latency connections) FC  special links and switches, lossless FCOE  Fiber channel over ethernet iSCSI  IP networks with normal Ethernet switches CIFS  windows based NFS  Linux based LAN Pin Groups Each vNIC- pins to link (all link active) To pin to particular link (to reserve bandwidth etc) When vNIC comes online, UCS use dynamic pinning Static pinning—created via LAN pin groups (pinning particular vNIC to particular uplink) but looses resiliency