Pre Training 3/26/2024

• Purpose-built all-flash array.
• Performance, data reduction, simplicity, space-efficient snapshots
• Commodity hardware
• Active/Active controllers (bricks)
• Scale-out design
• XDP = extreme data protection. Blends RAID 10 ,RAID 5 and RAID 6
• EMLC flash drives

• As well as compression
• Integrates with Vblock, VSPEX, VMware Vcenter or VAAI, App mobility/DR, Multipath Failover ( PowerPath)

• DARE

2 brick environment

• Restart module, R module, D module, C module. SW failure
• Reads journals, up-to-date state, lazy load process read metadata from drives
• Module failover, node bad. Another node of the brick. Whichever module.
• R module uses multi-pathing, Powerpath.
• If the node fails, all the journal memory is gone for that node. Journal memory reassignment to other nodes/controllers. If node is down or no longer has UPS protection.
• If no communication to UPS array assume it is bad. UPS tested every second.
• IB switches are not UPS-protected.
• Journal memory is considered non volatile, and must be able to dump to drives.
• Working to replace UPS protection with NVRAM instead. Which retains content if power is lost.
• Maybe XIO 1 uses UPS and XIO 2 uses NVRAM?
• During a power event, every node acts as itself and dumps the journal and shuts itself down. Don’t want to depend on the system manager. Takes less than a minute to do emergency shutdown, dump journal and shut down.
• Will notify SYM aka system manager. Usually running only on one node/controller.
• IPMI connectivity to nodes. A clustering agent can fence parts off
• 1 UPS per brick
• If both IB switches fail, the platform manager for each node reacts by performing an emergency shutdown. Will dump the journal but will stay powered on if UPS is good
• JBOD. 2 LCC cards, Each card has 2 SAS ports
• Node has 1 LSI card with 2 ports
• RDMA = remote direct memory access
• When UPS fails, it can no longer hold journal memory on the nodes protected on that UPS. Reassigned memory on other nodes.
• If only DAE loses power, SYM will close the gates until power is restored. Can’t serve IO

• XIOS = extreme I/O operating system. Running C in Linux
• Many to many system, everyone sees everything

• If you lose both nodes in the first brick you lose service. L and M modules. Clustering and SYM agents.
• Each volume can be up to 4 PB in size. 4K blocks
• Journal = a way of protecting the array before writing to SSD. In memorey faster and batches of writing to SSDs, much more efficient. Data and metadata. Delay writing to SSD, which only happens in background processes. Memory chunks on all the nodes in the system. Redudant