The Cloud Pak for Data System connector node is purpose-built to accelerate and improve the performance of your Cloud Pak for Data and Netezza Performance Server engine by enabling the use of more storage arrays such as SAN, NAS and other lower cost storage options.
Connector node is a Lenovo based, 2U server platform with integrated storage, processing and I/O. The chassis comes as a pretested building block loaded with software. It is installed and configured by IBM.
The chassis can be installed in a Cloud Pak for Data System rack or a customer supplied rack meeting the guidelines in the Customer Install Guide, Rack and System document included on the Safety and Regulatory CD shipped with your Cloud Pak for Data System components.
Since the base Cloud Pak for Data System and Netezza Performance Server nodes do not have any PCIe slots available for fibre channel HBAs, the connector node provides the ability to use fibre channel (FC) block storage for high-speed backup and restore, load, and unload for Netezza Performance Server.
Benefits of the solution
- Higher-speed backup and restore, load, and unload performance for Netezza Performance Server (NPS)
- Base Cloud Pak for Data System and NPS SPUs do not have any PCIe slots available for FC HBAs, connector nodes add that option.
- Faster backup can help with large systems where more data has to be backed up in the same windows of time as small systems
- Connector nodes provide high-performance backup and restore capability with 32 CPU cores, 192GB DDR4, 4 32Gb FC ports, 8 direct-attach NVMe drives, and 2 bonded 100Gb Ethernet ports to the internal fabric switches.
- When connector node is present on a system, the NPS host container runs on it, providing enhanced compute and memory resources.
- Connector node has 2x100GbE uplink to the fabric switch, meaning faster data load and unload over Ethernet with a lower CPU burden on the host.
- Backups can be done over FC (LAN-free) at a rate of 10.65TB/hour (compressed) using
zstd, for a single NPS instance. This was tested on a base+2 system with a very fast storage landing zone. Customer results might be better on larger systems, but worse if the customer storage is on the slower side.
dropdata(host writes to
/dev/nullinstead of remote storage over LAN or FC) speeds improve from 3.7TB/hour to over 10TB/hour, opening up an existing bottleneck internal to the system itself, which had been hit at some systems with fast enough storage and house network. This can help even with LAN-based backups for customers without FC storage connections.
- Two or more connector nodes can be used for failover:
- If the system has one connector node and it fails, access to the FC storage for backup and restore is lost until the node is fixed or repaired/replaced. The NPS host container fails back over to a control plane node, and the available CPU and memory is reduced back to a normal NPS installation.
- If there are two connector nodes and one fails, the host container fails over to the other connector node, and access to FC for backup and restore is not lost, and no performance degradation is experienced.
|Server type||SR650 Gen2|
|CPU||2 x 16C 2.1 GHz||Silver 4216 Processors|
|Total Memory||192 GB|
|Memory DIMM||16 GB 2933 MHz DDR4||2R x 8|
|LOM Slot||2 x 1G RJ-45||FC: AUKG|
|PCI Slots||DP 100G NIC||Front-end|
|2 x DP 32G FC HBA||Back-end|
|Internal Drives||8 x 3.84 TB NVMe||Toshiba CM5 (1 DWPD)|
|Internal M.2 HBA||Marvell 88SE9230||Part of M.2 Enablement Kit|
|Internal M.2 Drives||2 x 480 GB SATA||Micron 5100 (1.5 DWPD)|
|Power Supply||2 x 1,100 W||Fully redundant|
Connector node setup in the system
- Cloud Pak for Data System of size Base+8 or greater and including NPS comes with two connector nodes by default. Smaller systems may optionally be configured with one connector node, however, in this setup the ability to backup and restore to/from FC connections is not HA.
- In the systems with two connector nodes, only one connector node is active at a time, the other is for high availability.
- The system provides the ability to switch the connector node personality between UNSET and NPS Host.
- The NPS host container runs on bare metal on connector node.
- FC-based backup and restore, load, and unload for NPS via connector node
Connector nodes can be optionally used for filesystem-based backup, or using third-party LAN-free backup such as TSM/Spectrum Protect-based backup LAN-free. For more information, see https://www.ibm.com/docs/en/netezza?topic=spftsmc-setting-up-tivoli-storage-manager-8110-lan-free.
- If using FC, 4 ports per connector node are required on the customer SAN.
- Customers might back up to spinning disk or to tape. These can be a bottleneck after adding the connector node. Consider many HDDs in parallel (can be RAIDed or not – the connector node can handle both/either). Benchmark the raw storage with fio/iozone or similar for reference.
- Customer storage and TCP/IP network (or FC SAN) must be able to support the I/O write bandwidth that corresponds to the targeted backup speed. (8Gb/s = 1GB/s = 3.6TB/hour).
- Cloud Pak for Data System does not support adding a customer server to the same GPFS cluster as the one internal to the system. Therefore, if customers must use TSM with directory-container storage pools, they must stick with LAN-based backup. However, LAN-based backups can still benefit with connector node. For example, in the case of 2x10Gb/s bonded Ethernet house connection., up to 2.5GB/s (9TB/hour) is the limit. Of course, backups would contend for network bandwidth with other traffic, and network bond efficiency is not perfect. The result between 3.7TB/hour and 9TB/hour could be achieved with connector node with LAN-based backup (the backup rate between 3.7TB/hour and 9TB/hour is specific for TSM LAN-based only).
- FC block storage provided to connector node must be clear of any filesystem as it will be overwritten!
- If using FC connections, customer has to add the WWPNs to their hosts for the desired LUNs, and make sure to exclude other LUNs from being exposed to those WWPNs. This is important because the software will overwrite all LUNs it has access to (customer will be prompted first).
nzbackupresults are fastest using
zstd. NPS 11.2 adds ability for
zstdformat to be restored to PDA systems. It is advised to use this format. Moreover, consider proper number of backup sessions and streams.