Server Sizing

On-premises configurations

500 seats to 50,000 seats

500 seats to 250,000 seats

Single Server Config

tachyon_singleserver_sizing

Split Server Config

tachyon_serversizing_splitconfig

	Small 1	Small 2	Small 3	Small 4	Medium 1	Medium 2
Maximum number of devices	500	2,500	5,000	10,000	25,000	50,000
Tachyon Server
CPU cores	2 Cores	3 Cores	3 Cores	5 Cores	6 Cores	12 Cores
RAM	8 GB	10 GB	16 GB	18 GB	22 GB	32 GB
SQL Instance Max Memory	1 GB	1 GB	2 GB	2 GB	4 GB	8 GB
Number of Tachyon Switches	1	1	1	1	1	1
Network Interfaces	1	1	1	1	1	1
Disk space required for Database	0.5 GB	1.5 GB	3 GB	4 GB	15 GB	52.5 GB
SQL Server Disk space requirements breakdown
Tachyon Master Database MDF	183 MB	183 MB	398 MB	398 MB	398 MB	398 MB
Tachyon Master Database LDF	8 MB	8 MB	72 MB	72 MB	72 MB	72 MB
TempDB MDF	64 MB	64 MB	64 MB	64 MB	64 MB	64 MB
TempDB LDF	8 MB	8 MB	8 MB	8 MB	8 MB	8 MB
Tachyon Responses DB requirements breakdown
Number of concurrent questions	10	20	25	25	50	100
Total Response DB storage per device	0.1 MB	0.2 MB	0.2 MB	0.2 MB	0.5 MB	1 MB
Tachyon Response Database MDF	50 MB	500 MB	1 GB	2 GB	12.5 GB	50 GB
Tachyon Response Database LDF	50 MB	400 MB	1 GB	1.25 GB	1.5 GB	1.5 GB

	Small 1	Small 2	Small 3	Small 4	Medium 1	Medium 2	Large 1	Large 1.5	Large 2	Large 2.5
Maximum number of devices	500	2,500	5,000	10,000	25,000	50,000	100,000	150,000	200,000	250,000
Tachyon Server
CPU Cores	1 Core	2 Cores	2 Cores	3 Cores	4 Cores	8 Cores	12 Cores	16 Cores	20 Cores	24 Cores
RAM	6 GB	8 GB	12 GB	14 GB	16 GB	24 GB	42 GB	60 GB	78 GB	96 GB
Number of Tachyon Switches	1	1	1	1	1	1	2	3	4	5
Network Interfaces	1	2	2	2	2	2	3	4	5	6
Database Server
CPU Cores	1 Core	1 Core	1 Core	2 Cores	2 Cores	4 Cores	5 Cores	6 Cores	7 Cores	8 Cores
RAM	4 GB	4 GB	6 GB	6 GB	8 GB	12 GB	16 GB	20 GB	24 GB	28 GB
SQL Instance Max Memory	1 GB	1 GB	2 GB	2 GB	4 GB	8 GB	12 GB	16 GB	20 GB	24 GB
Disk space required for Database	0.5 GB	1.5 GB	3 GB	4 GB	15 GB	52.5 GB	153 GB	228.5 GB	404 GB	504 GB
SQL Server Disk space requirements breakdown
Tachyon Master Database MDF	183 MB	183 MB	398 MB	398 MB	398 MB	398 MB	415 MB	415 MB	415 MB	415 MB
Tachyon Master Database LDF	8 MB	8 MB	72 MB	72 MB	72 MB	72 MB	328 MB	520 MB	520 MB	520 MB
TempDB MDF	64 MB	64 MB	64 MB	64 MB	64 MB	64 MB	64 MB	64 MB	64 MB	64 MB
TempDB LDF	8 MB	8 MB	8 MB	8 MB	8 MB	8 MB	8 MB	8 MB	8 MB	8 MB
Tachyon Responses DB requirements breakdown
Number of concurrent questions	10	20	25	25	50	100	200	200	250	250
Total Response DB storage per device	0.1 MB	0.2 MB	0.2 MB	0.2 MB	0.5 MB	1 MB	1.5 MB	1.5 MB	2 MB	2 MB
Tachyon Response Database MDF	50 MB	500 MB	1 GB	2 GB	12.5 GB	50 GB	150 GB	225 GB	400 GB	500 GB
Tachyon Response Database LDF	50 MB	400 MB	1 GB	1.25 GB	1.5 GB	1.5 GB	2 GB	2.25 GB	2.75 GB	3 GB

Tachyon Switch Server Sizing

The recommended configuration for the Tachyon Switch when hosted separately on its own server is 2 CPU Cores and 16 GB RAM.

Benchmarking Hardware

Benchmarked on Windows Server 2012 R2 Hyper-V infrastructure, with Database and Application Server components setup on separate VMs hosted on same physical host.
CPU: Hyper-V Host CPU configuration - 2x Intel Xeon CPU E5-2867W v3 @ 3.10 GHz, 25MB Cache, 10C/20T, Max Mem 2133MHz.
Network: Application server and Database server were connected over a 10Gbps link through a 10Gbps physical switch. External connections for the VMs were over a 1Gbps link through a 1Gbps physical switch.
Database storage:
- Intel DC P3700 Solid State Drives (SSDs) with up to 450K/175K IOPS (4K random read/write QD32), and Samsung 850 PRO SSDs with up to 100K/90K IOPS (4K random read/write QD32), attached locally to the Hyper-V host.
- MDF, LDF and TempDB files on separate disks, with MDF on Intel DC P3700 SSDs, and LDF and TempDB on Samsung 850 PRO SSDs.

Other Recommendations

Physical or VM

The servers can be deployed either on physical or virtual machines. For deployment on a virtual machine, the CPU cores should be assigned at 100% virtual machine reserve.

Network

For a remote (split) server configuration, the Tachyon Server requires a dedicated network interface (MAC Address and IP Address) for connection to the remote SQL Server instance used for the Responses database.
This dedicated SQL connection must be at least 1Gbps for Lab, Small and Medium environments (< 100000 devices) and at least 10Gbps for Large environments (> 100000).

Database Server

Split Data, Log and TempDB on separate disks for medium and large configurations.
SQL Server must be configured with maximum server memory limit and not left at the default setting to consume unlimited memory.
For sizing the Database server in the recommendations above, 4GB RAM has been added for the operating system on top of SQL Server instance RAM requirements.

Tachyon Server

In previous versions of Tachyon, you may have had to increase the UploadBatchSize value to 600 (default 300) in the Master database SwitchConfiguation table. This version of Tachyon does not require this configuration. If you changed this value in a previous version and have upgraded to the latest version of Tachyon, then you do not need to reset or change the value again.

Cloud configurations

The following tables describe the supported instance sizing for various numbers of supported clients for both single server and split server implementations, as per on-premises sizing.

When compared with the on-premise sizing guidelines these instance sizes will generally provide a greater number of CPU cores and/or memory. There is an overhead inherent in commodity virtualized cloud computing solutions which prevents them from providing exactly the same level of performance as dedicated on-premise virtualization solutions. In some cases it is also necessary to provide more of a particular resource type than required in order to get sufficient levels of another resource type due to the limited number of instance configurations offered by the cloud provider.

Instance sizing is subject to change, and may not be available in all geographic regions.

AWS	Lab	Up to 500	Up to 2.5K	Up to 5K	Up to 10K	Up to 25K	Up to 50K
Tachyon with local SQL Server	t2.medium	m5.large	m5.xlarge	m5.xlarge	m5.2xlarge	m5.2xlarge	m5.4xlarge
Tachyon Server	t2.small	m5.large	m5.large	m5.xlarge	m5.xlarge	m5.xlarge	m5.2xlarge
Remote SQL Server	t3.medium	r5.large	r5.large	r5.large	r5.large	r5.xlarge	r5.2xlarge

Azure	Lab	Up to 500	Up to 2.5K	Up to 5K	Up to 10K	Up to 25K	Up to 50K
Tachyon with local SQL Server	A2_V2	A4_V2	D3_V3	DS12_V2	DS4_V2	DS4_V2	F16S
Tachyon Server	A1_V2	D2_V3	D11_V2	D11_V2	D4_V3	D12_V2	DS4_V2
Remote SQL Server	A2_V2	DS2_V2	DS2_V2	DS2_V2	DS2_V2	DS2_V2	DS3_V2

AWS Network Configuration

For a split server installation it is necessary to configure an additional network interface into each server to guarantee a dedicated 1Gbps network channel between the core and the database. Enhanced networking (SR-IOV) must be enabled on both of these NICs which should be the default. Detailed steps on how to verify this are given here - http://docs.aws.amazon.com/AWSEC2/latest/WindowsGuide/sriov-networking.html. Also increase the transmit (TX) and receive (RX) buffer sizes to their maximum under the network adaptor advanced properties.

Azure Network Configuration

For a split server installation it is necessary to configure an additional network interface into each server to guarantee a dedicated 1Gbps network channel between the core and the database. Enhanced networking (SR-IOV) must be enabled on both of these NICs where this is available (only on instances with 8cores or more) and must be enabled during instance creation. The additional NIC must also have accelerated networking enabled using Azure Powershell or Azure CLI. Detailed steps on how to configure this are given here - https://docs.microsoft.com/en-us/azure/virtual-network/virtual-network-create-vm-accelerated-networking. Also increase the transmit (TX) and receive (RX) buffer sizes to their maximum under the network adaptor advanced properties within the guest OS.

If connecting to an Azure based Tachyon switch via external public Azure IP addresses or Azure Load Balancers you will need to extend the default TCP timeout from 4 seconds to 15 seconds - Detailed steps on how to do this can be found here - https://azure.microsoft.com/en-us/blog/new-configurable-idle-timeout-for-azure-load-balancer/

AWS Database Storage Configuration

Database instances (either combined or the database server in a split implementation) should use an additional Throughput Optimized SSD (ST1) EBS volume to guarantee sufficient disk performance. This is used for all database storage - data, log and tempdb. The specified instances do not support any more than one EBS volume. If you wish to separate data, log and tempdb onto separate volumes (as per Microsoft recommendations) you will need a higher spec instance for the DB server, one that will allow you to attach more volumes and allow higher throughput to disk.

For any implementation other than Lab the Elastic Block Storage (EBS) optimization must be enabled, either when the instance is originally launched or afterwards. Details on how to do this can be found here: http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/EBSOptimized.html.

Azure Database Storage Configuration

Database instances (either combined or the database server in a split implementation) should have a minimum of 2 SSDs for database files. One for the Tachyon data files and tempdb and the other for the log files.

Disable caching on the volume used for the SQL log drive. This is done on the Azure console and not within the guest. Remember to restart the DB server VM once this is changed.

Do not use geo-redundant storage for any of the SQL drives.

A full list of optimizations for running SQL on Azure is available here: https://docs.microsoft.com/en-us/azure/virtual-machines/windows/sql/virtual-machines-windows-sql-performance

Page tree

Publish this draft for everyone to see.

Other users have edits for this page:

Permanently discard changes to this page.

This cannot be undone.

Contents