Plan for Bare Metal Solution

This topic explains the key components of Bare Metal Solution to help you decide how to deploy servers, networking, and storage in your own Bare Metal Solution environment.

If you are running SAP HANA in your Bare Metal Solution environment, then see the SAP HANA on Bare Metal Solution planning guide.

Regional availability

Bare Metal Solution is available in the following Google Cloud regions:

Region Location
asia-northeast1 Tokyo, Japan, APAC
asia-northeast3 Seoul, South Korea, APAC
asia-southeast1 Jurong West, Singapore, APAC
europe-west2 London, England, Europe
europe-west3 Frankfurt, Germany, Europe
europe-west4 Eemshaven, Netherlands, Europe
europe-west6 Zurich, Switzerland, Europe
northamerica-northeast1 Montréal, Québec, Canada, North America
northamerica-northeast2 Toronto, Ontario, Canada, North America
southamerica-east1 Osasco, São Paulo, Brazil, South America
us-central1 Council Bluffs, Iowa, USA, North America
us-east4 Ashburn, Virginia, USA, North America
us-west2 Los Angeles, California, USA, North America

The following table shows the current Google Cloud regions that support SAP HANA on Bare Metal Solution.

Region Location
europe-west3 Frankfurt, Germany, Europe
europe-west4 Eemshaven, Netherlands, Europe
us-central1 Council Bluffs, Iowa, USA, North America
us-east4 Ashburn, Virginia, USA, North America

If you do not see the region that you need in the preceding table, contact Google Cloud Sales.

Server configurations

Bare Metal Solution offers servers in pre-configured sizes, as shown in the following table.

Each bare-metal server comes with 200 GB of network attached storage for the boot disk.

By default, CPU hyper-threading is disabled in Bare Metal Solution servers. If you need hyper-threading, let your sales representative know or, if you already have Bare Metal Solution, open a service request.

Server name CPU cores vCPUs Sockets Memory CPU platform Ethernet ports
General-purpose servers
o2-standard-16-metal 8 16 2 192 GB Intel Xeon Gold, 5200 series, 3.8 GHz 4 port 25GbE NIC
o2-standard-32-metal 16 32 2 384 GB Intel Xeon Gold, 6200 series, 3.2 GHz 4 port 25GbE NIC
o2-standard-48-metal 24 48 2 768 GB Intel Xeon Gold, 6200 series, 3.0 GHz 4 port 25GbE NIC
o2-standard-112-metal 56 112 2 1.5 TB Intel Xeon Platinum, 8200 series, 2.2 GHz 4 port 25GbE NIC
High-memory general-purpose servers
o2-highmem-224-metal 112 224 4 3 TB Intel Xeon Platinum, 8200 series, 2.7 GHz 4 port 25GbE NIC
Memory-optimized servers
o2-ultramem-672-metal 336 672 12 18 TB Intel Xeon Platinum, 8200 series, 2.7 GHz 6 port 25GbE NIC
o2-ultramem-896-metal 448 896 16 24 TB Intel Xeon Platinum, 8200 series, 2.7 GHz 8 port 25GbE NIC

Firmware

We recommend that you install the latest version of Bare Metal Solution firmware on your servers: TS54.

Using an outdated firmware version can cause issues including the following:

  • Delayed and limited Google Cloud support.
  • Loss of critical Google Cloud features.
  • Security, compliance, and performance issues.

To learn how to upgrade firmware, see Upgrade firmware.

Identify your firmware version

To identify the current firmware version of your Bare Metal Solution server, run the following command:

sudo dmidecode -s bios-version

Available firmware versions

The following table provides details for the firmware versions available for Bare Metal Solution.

General information

Firmware version Machine series Machine type Lifecycle stage EOL and firmware deprecation date
BIOS_PUR043.45.00.002 (TS54)
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose GA To be determined
BIOS_PUR043.37.14.021 (TS24.02)
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose GA April 2024
BIOS_PUR043.37.16.023 (TS24.05)
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose GA April 2024
3.50.58
  • o2-ultramem-672-metal
  • o2-ultramem-896-metal
Memory-optimized servers GA March 2024
3.80.24
  • o2-ultramem-672-metal
  • o2-ultramem-896-metal
Memory-optimized servers GA To be determined

Features

Firmware version Features
BIOS_PUR043.45.00.002 (TS54)
  1. Support for interactive serial console.
  2. Reliability, availability, and serviceability (RAS) features:
    • Adaptive Double Device Data Correction (ADDDC)
    • Post Package Repair (PPR) to reduce DIMM errors, repair DIMM issues, reduce maintenance downtime
  3. Better reliability when a server shuts down or starts up.
  4. Bug fixes for the following:
    • Unified Extensible Firmware Interface (UEFI) security issue
    • Reboot issues linked to loading the initial RAM disk
    • Baseboard Management Controller (BMC) issues related to the IPMI internal client
    • False alarms for fan status and CPU voltage

Upgrade firmware

To upgrade your server to the latest firmware version, contact Customer Care.

We need a 5-day notice to schedule the upgrade. The upgrade requires approximately 10 hours to complete. We can upgrade three servers concurrently in a 10 hour window.

Interactive serial console

Bare Metal Solution provides an interactive serial console to access your servers. You can use it to perform configuration and troubleshooting tasks. You can directly run commands and respond to prompts in the serial console.

To learn how to configure and use the serial console, see Configure the serial console.

Operating systems

When you place an order for Bare Metal Solution, you can specify the operating system (OS) that you need installed on your servers. If you need an OS that does not appear in the following tables, contact Google Cloud Sales.

Linux

You can request the installation of the following Linux OS vendor versions on your Bare Metal Solution servers. This section provides general operating system (OS) details for the OS images.

OS version OS image code Machine series Machine type Lifecycle stage EOL and image deprecation date
RHEL 7.7 for SAP LAMBRHEL77SAP
  • o2-ultramem-896-metal
Memory-optimized servers (appropriate for SAP HANA) End of Life August 2023
RHEL 7.9 RHEL7x
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose GA June 2024
RHEL 8.4 (Minor version locked) RHEL84
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose End of Life May 2023
RHEL 8 (Latest minor release) RHEL8x
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose GA May 2029
RHEL 8.0 for SAP RHEL80SAP
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose End of Life December 2020
RHEL 8.1 for SAP RHEL81SAP
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose End of Life November 2023
OEL 7.9 OEL7x
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose GA July 2024
OEL 8 (Latest minor release) OEL8x
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose GA July 2029
SLES 15 SP1 SLES15SPx
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
General-purpose End of Life January 2024
SLES 12 SP4 for SAP SLES12SP4SAP
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
  • o2-ultramem-896-metal
General-purpose & Memory-optimized servers (appropriate for SAP HANA) End of Life June 2023
SLES 12 SP5 for SAP SLES12SP5SAP
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
  • o2-ultramem-896-metal
General-purpose & Memory-optimized servers (appropriate for SAP HANA) GA October 2027
SLES 15 for SAP SLES15SAP
  • o2-standard-48-metal
  • o2-standard-32-metal
  • o2-standard-112-metal
  • o2-standard-16-metal
  • o2-highmem-224-metal
General-purpose End of Life December 2022
SLES 15 SP1 for SAP SLES15SPxSAP
  • o2-standard-16-metal
  • o2-standard-32-metal
  • o2-standard-48-metal
  • o2-standard-112-metal
  • o2-highmem-224-metal
  • o2-ultramem-896-metal
General-purpose & Memory-optimized servers (appropriate for SAP HANA) End of Support January 2024
SLES 15 SP2 for SAP LAMBSLES15SP2SAP
  • o2-ultramem-896-metal
Memory-optimized servers (appropriate for SAP HANA) GA December 2024

Hypervisors

When you place an order for Bare Metal Solution, you can request that a hypervisor be installed on your servers. If you request a hypervisor, you must install any guest operating system yourself.

You can request the following hypervisor:

  • Oracle VM
    • 3.4.6

To use Oracle Linux Virtualization Manager (OLVM - a KVM-based hypervisor), request a compatible Linux OS such as Oracle Linux 7.

Storage

For storage in a Bare Metal Solution environment, you need to select the type of storage you need, know how to maximize the performance, and decide if you want to enable snapshots.

Storage choices

Storage devices in the regional extension provide local storage to Bare Metal Solution servers. There are three types of local storage you can request:

  • Fibre Channel-based block storage presented as logical unit numbers (LUNs) and accessible by only one Bare Metal Solution server.
  • Fibre Channel-based block storage presented as multipath LUNs and shared with multiple Bare Metal Solution servers. This option is good for high-availability applications, such as Oracle Database RAC.
  • Network File System (NFS) storage. With NFS, you mount a shared storage directory on one or more Bare Metal Solution servers and access the directory like any other file system. For information on how to implement NFS, see Add NFS file storage to a Bare Metal Solution environment.

For Fibre Channel-based block storage, you can choose the following storage options in 1 TiB increments up to a maximum of 16 TiB. For NFS, you can choose the following storage options with a minimum of 100 GiB and a maximum of 16 TiB.

Storage type Snapshot support Common Workloads
SSD (solid-state drive) Yes Mission-critical production workloads
HDD (hard drive) Yes Local backups or archival workloads

Here are some additional storage volume considerations that you need to understand:

  • We don't recommended using hard disk drives (HDDs) to run mission-critical Oracle production workloads, because performance guidance is not available for HDD storage.
  • When you provision any Fibre Channel or NFS storage volume, the volume size becomes fixed and you cannot expand the volume.
  • If you need additional storage space, you need to purchase additional storage volumes.
  • You cannot expand the existing LUNs in a storage volume. Instead, you need to order new LUNs and a new storage volume, migrate the old LUNs to the new storage volume, and decommission the old LUNs.
  • When you migrate the LUNs to a new storage volume, you also need to migrate the data.
  • If you run Oracle Automatic Storage Management (ASM), the following options are available:

    • When you need to expand an ASM disk group, you can order another storage volume, rescan the SCSI hosts to discover the new LUNs, and use ASM to add the new storage capacity into the disk group.

      You can use the following command to perform an SCSI rescan on Red Hat Linux or Oracle Linux:

      sudo rescan-scsi-bus.sh -a
      
    • When you add, drop, or resize disks, ASM automatically initiates a rebalance to prevent an unnecessary load on the production database.

  • When you remove a storage volume from service in a Bare Metal Solution environment, the Bare Metal Solution server reboots. The length of the outage can last from 20 minutes to an hour, depending on the size of your Bare Metal Solution server and the amount of storage attached to the server.

Storage performance

For Fibre Channel and NFS storage, Bare Metal Solution provides throughput and input/output operations per second (IOPS) performance calculated according to the following formula for an 8 KiB block size on an SSD drive:

Calculated IOPS = 7200 * Volume Size in TiB
Maximum Throughput (MiBps) = Calculated IOPS * 8 / 1024

The following table shows the IOPS and the throughput for common volume sizes:

Volume size Read/Write IOPS Throughput (MiBps) Read/Write latency
200 GiB (0.2 TiB) 1400 10.9 < 1 ms
1 TiB 7200 56 < 1 ms
4 TiB 28800 225 < 1 ms
8 TiB 57600 450 < 1 ms
12 TiB 86400 675 < 1 ms
16 TiB 115200 900 < 1 ms

Use the following storage recommendations to improve the performance of your databases and applications in your Bare Metal Solution environment:

  • When you deploy multiple storage volumes, use consistent volume sizes to get consistent performance for databases and applications.
  • We recommend that you configure 1 LUN for each volume with 4 TiB or less, and 8 LUNs for volumes with more than 4 TiB.
  • When you add a new storage volume, always remember to rescan the LUNs to detect the new storage capacity.

Storage volume snapshots

The Bare Metal Solution provides options to save snapshots of your storage volumes to use for recovery purposes. Snapshots are available as follows:

  • OS boot volume: Google Cloud takes a daily snapshot by default.
  • Data volume: You can request daily snapshots when you place your initial order for Bare Metal Solution. You can also contact Customer Care to request data volume snapshots for existing storage volumes.

Google Cloud takes snapshots of the boot OS volume automatically every day at 12:10 AM in the region where your Bare Metal Solution servers are located. We also take a snapshot of your data volume if you have ordered this. We retain the snapshots for 7 days, and you cannot modify the schedule for these snapshots.

The following diagram shows the default process of how snapshots fill up a storage volume.

Visual representation of how snapshots fill up a storage volume and need to be
deleted to allow new snapshots

  • When Google Cloud enables snapshots on a data volume, we reserve 80% of the volume for regular data and 20% of the volume for snapshots. The latter is called the snapshot reserve.
  • When the snapshot reserve fills its original 20% space, the volume grows automatically by an additional 20%. This increases both the data area and the snapshot reserve proportionally.
  • If the snapshot reserve becomes 100% full, snapshots spill over into the data area.
  • If both the data area and snapshot reserve become 100% full, the volume goes offline to prevent corruption of the LUNs. Write operations are suspended.
  • To bring the volume back online, you need to reclaim sufficient space in the volume by deleting prior snapshots.

Your data volume might fill up as the result of a high rate of data change within the LUN(s). For example, an application might write and delete large amounts of data within a 24-hour period.

To delete data volume snapshots and bring a data volume back online, or restore an OS boot volume snapshot, contact Customer Care.

Networking

Each Bare Metal Solution environment resides in a secure region extension that is connected to the corresponding Google Cloud region by a Partner Interconnect connection that is provided and managed by Google Cloud.

Google Cloud monitors round-trip latencies between VM instances in Compute Engine zones and the Bare Metal Solution environment in the same region. For Google Cloud locations in Los Angeles, Ashburn, and London, median round-trip latencies are regularly measured at 0.6 ms to 1 ms.

Choose standard networking or explore advanced networking options

The standard networking configuration for Bare Metal Solution offers a basic network topology that is appropriate for most small-scale deployments. The standard networking configuration includes:

  • Four physical interfaces for each Bare Metal Solution server. We bundle the interfaces into two logical interfaces called bonds. These bonded interfaces use a standard networking template to provide basic redundancy and load balancing.
    • One bonded interface connects to a single, client VLAN network to reach Google Cloud.
    • The second bonded interface connects to a single, private VLAN network for services and applications running locally in your Bare Metal Solution environment.
  • A single Partner Interconnect connection with a single pair of VLAN attachments connects your Google Virtual Private Cloud to the client VLAN in your Bare Metal Solution environment.

We recommend using the following Bare Metal Solution default network configuration for small scale deployments:

Small Bare Metal Solution default network configuration.

It consists of the following:

  • A single client VLAN attachment connected to the Google Cloud using the Partner Interconnect.
  • A single private VLAN attachment only connecting the Bare Metal Solution servers.
  • A single Partner Interconnect (pair of VLAN attachments).

If the standard networking configuration does not meet your needs, you can consider implementing some advanced networking capabilities. If you already have a Bare Metal Solution environment and would like to migrate to an advanced networking configuration, contact Customer Care. If you are a new customer and would like to implement an advanced networking configuration, contact Google Cloud Sales.

When you order an advanced networking configuration, the following options become available for more complex network deployments:

Multiple VLANs

Instead of a single VLAN for each bonded interface, you can add multiple VLANs on the same bonded interface.

  • The additional VLANs can be client VLANs, private VLANs, or a combination of both.
  • You can add up to 10 VLANs per bonded interface and a total of 20 VLANs per server.
  • You can enable routing between different client VLANs in the same networking environment.

  • To add multiple VLANs to your Bare Metal Solution servers, you have two options:

    • Request a server reimage and specify which VLANs you want added to the server bonded interfaces.

    • Request a network change, then use your operating system to add the selected networks to the server bonded interfaces using information received from Google Cloud Support.

Multiple VLAN attachments

You can configure multiple VLAN attachments over a Partner Interconnect connection to your Bare Metal Solution environment.

  • This provides more bandwidth between your VPC and the Bare Metal Solution environment than the bandwidth offered by a single set of VLAN attachments.
  • The network load balances the traffic across the multiple VLAN attachments.
  • Based on network conditions, total available bandwidth might be lower than the sum of the bandwidth of the Partner Interconnect connections.
  • To configure multiple VLAN attachments, see Create the VLAN attachments for the Partner Interconnect connection.

Multiple VPC connections

You can connect the Bare Metal Solution environment to more than one Virtual Private Cloud (VPC) in your Google Cloud project.

Maximum transmission unit

You must use the same maximum transmission unit (MTU) for every VLAN to ensure consistent switching between devices.

  • Select either a standard frame size (1500 bytes) or jumbo frames (9000 bytes).
  • Specify the same MTU for every VLAN in the client VLAN network and the private VLAN network.
  • Configure the MTU on your Bare Metal Solution servers by using your server's operating system. For example, to configure jumbo frames on a Linux server, include the ip link set INTERFACE_NAME mtu 9000 command on all server interfaces.
  • To verify the current MTU settings in a Linux server, issue the ip link show | grep mtu command.
  • Do not change the MTU setting on your VLANs after initial provisioning.

Network templates

You can use network templates to enable a flexible network design of your choice, such as active/active and active/backup, to meet your needs for high availability, redundancy, and load balancing.

  • Templates allow for different combinations of connections between server interfaces, bonds, and switch ports.
  • To learn more about templates and request them for your Bare Metal Solution network, contact Google Cloud Sales.

Partner Interconnect for Bare Metal Solution

The connection between your Google Cloud environment and your Bare Metal Solution environment uses the Partner Interconnect framework.

For more information about the Partner Interconnect framework, see the Partner Interconnect documentation, but keep in mind the following differences applicable in the context of Bare Metal Solution:

  • Because Google Cloud manages the network connection, you provide pairing keys for the VLAN attachments to Google Cloud instead of a third-party service provider.
  • This connection to the Bare Metal Solution environment is not subject to the pricing model that the Partner Interconnect product uses. For more information about the pricing model, talk to your Google Cloud Sales representative.

Define your own IP address spaces

Bare Metal Solution enables you to bring your own IP subnets for use in the bare-metal environment. This enables you to fit your Bare Metal Solution into any existing IP address schema that your enterprise might have.

When you place an order for Bare Metal Solution, you provide two internal IP address ranges: a client IP address range for communication between your Google Cloud and Bare Metal Solution environments, and a private IP address range used for accessing services and devices within the Bare Metal Solution environment.

You can also select the specific IP addresses to use with each server. Consider the following guidelines:

  • Make sure that no existing IP addresses in your VPC network or your extended on-premises network conflict with the IP ranges that you use for Bare Metal Solution environment.
  • For every subnetwork that you request for Bare Metal Solution, Google Cloud reserves five IP addresses from the subnet range for administration purposes. These include the network address, the broadcast address, and the last three usable addresses in the range. For example, if you request the 100.64.0.0/27 subnet range, we reserve the following addresses:

    • 100.64.0.0: Network address
    • 100.64.0.28, 100.64.0.29, and 100.64.0.30: Last three usable addresses in the range.
    • 100.64.0.31: Broadcast address
  • Request subnet ranges with at least a /27 subnet, so that you can assign the /28 subnet for Bare Metal Solution services. Subnets using /29 to /32 do not contain enough usable addresses for either Bare Metal Solution services or your Bare Metal Solution servers.

  • In an auto mode VPC network, 10.128.0.0/9 CIDR block is reserved. See Auto mode IPv4 ranges. If you want to connect to an auto mode VPC network, use IP ranges outside of 10.128.0.0/9 for your Bare Metal Solution environment.

  • Bare Metal Solution does not support Class E IP addresses (240.0.0.0 to 255.255.255.254).

Multi-region networking considerations

For any outside communication, including communication with services or your on-premises environments, the communication from your Bare Metal Solution environment must pass through the VPC network that you are using to connect to your Bare Metal Solution environment.

Consider using a single VPC network with global_dynamic routing enabled if your Bare Metal Solution environment will require access to other Google Cloud regions to connect to any of the following resources:

  • Other Bare Metal Solution environments
  • VMs in other regions
  • Other Google Cloud services
  • Your on-premises sites in other geographical locations

Using a VPC network with global dynamic routing enabled and custom route advertisements, you can set up a routed environment to connect your Bare Metal Solution instances directly to your Google Cloud and extended on-premises environments in other regions, without any additional configuration in Google Cloud.

Our recommendation is to advertise a default route on the BGP sessions to Bare Metal Solution. This enables you to:

  • Avoid having to make change to routes to the Bare Metal Solution environment even if things change in your VPC subnets or your on-premises environments.
  • Keep the routing table to Bare Metal Solution small and allow it to contain only one prefix.

Network services and your Bare Metal Solution environment

As described earlier, Google Cloud recommends that you advertise a single default route to your Bare Metal Solution environment to keep the routing table simple. However, if your circumstances don't allow that, the routes that you need to expose to your Bare Metal Solution environment are determined by networks and services in your Google Cloud or on-premises environments that need access to the devices or applications in your Bare Metal Solution environment and vice-versa.

Some services that might require routes include NTP servers, DNS servers, LDAP servers, logging services, monitoring services, and so forth, as well as any other workloads and applications that need to communicate with the Bare Metal Solution environment.

Bare Metal Solution does not come with internet access. If you need to, you can setup routing to access the internet through any existing internet gateways that you have, either on Google Cloud or on premises.

Keeping routing simple

Outside of the other servers in the same Bare Metal Solution region, your servers likely need to communicate with one or more of the following:

  • Your endpoints in your Google Cloud environment
  • Your extended on-premises environment through your Google Cloud environment
  • Your servers in Bare Metal Solution environments in other regions through your Google Cloud environment
  • The internet through your Google Cloud environment

Firewalls and the Bare Metal Solution environment

When planning your firewall implementation for the Bare Metal Solution environment, there are a couple things to consider.

If you need to control traffic between the servers or applications within the Bare Metal Solution environment (east-west traffic), ask your Google Cloud sales representative for the available options.

In your VPC network on Google Cloud, you need to create firewall rules that allow incoming connections from the Bare Metal Solution environment.

Required networking information for your Bare Metal Solution order

When you submit a Bare Metal Solution order, include the following networking information:

  • The Google Cloud project ID that you are using with your Bare Metal Solution environment.
  • The IP ranges that you intend to use in your Bare Metal Solution environment. For example, the client IP address range used for communication between your Google Cloud and Bare Metal Solution environments, and the private IP address range used for services and devices within the Bare Metal Solution environment.
  • The number of VLANs you need in your Bare Metal Solution environment. Keep in mind that you need at least 2 VLANs for each Partner Interconnect.
  • The bandwidth, in 1 Gbps increments, that you need between your Google Cloud environment and your Bare Metal Solution environment.