H17000

H17000 is the 2U TR7 hardware platform running on two 3rd generation Intel Xeon Scalable processors. The values on this page are given with the workloads and metric definitions defined on the Test Methodology page; the condition of each value is stated in the related section.

The TR7 software runs the cryptographic operations within TLS and the data compression workloads together with the hardware acceleration capabilities offered by Intel processors and chipsets. Processor-level acceleration instructions and the acceleration resources supported by the platform are integrated into the processing flow of the software, enabling these operations to be performed more efficiently. The acceleration mechanisms used vary according to the workload and configuration. The models in the series share the same hardware platform.

200 Gbps

Throughput

The highest value in the series, H17200

7 M

HTTP requests/s

Over open connections, L7 service only

220,000

TLS TPS (ECDSA-P256)

Full handshake, ECDSA P-256

160 Gbps

TLS throughput

Bulk encryption on an established TLS connection

100 Gbps

Compression throughput

Pre-compression (input) bytes

50 M

L4 HTTP RPS

H17120 - H17200; L4 Concurrent Connections up to 300 M

Summary

Platform performance	H17080	H17120	H17140	H17160	H17200
Throughput	80 Gbps	120 Gbps	140 Gbps	160 Gbps	200 Gbps
HTTP requests/s	5 M	6 M	6 M	7 M	7 M
TLS TPS (ECDSA-P256)	140,000	200,000	200,000	220,000	220,000
TLS throughput	80 Gbps	120 Gbps	120 Gbps	140 Gbps	160 Gbps
Compression throughput	70 Gbps	80 Gbps	80 Gbps	100 Gbps	100 Gbps
L4 HTTP RPS	25 M	50 M	50 M	50 M	50 M
L4 Concurrent Connections	120 M	300 M	300 M	300 M	300 M

Metric definitions: Metric Definitions. Value classes (measured and capacity): Value Classes.

Platform

Property	Value
Processor	2 × 3rd generation Intel Xeon Scalable
Onboard network	2 × GbE RJ-45 (Intel I210)
Expansion	8 network module slots (highest option 2 × 100 GbE QSFP28 / slot)
Storage	2 × 1 TB SSD (RAID 1)
Power	1+1 redundant, 1300 W
Form factor	2U · 88 × 430 × 663 mm

The values are for the structure in which the network module slots are distributed across the two processors: each processor handles the traffic of its own network modules. When all network modules are attached to a single processor, the L7 values read about 17% lower.

L7 HTTP

Metric	Condition
HTTP requests/s	Kept-open HTTP/1.1 connections; TR7 generates the response itself, no Backend Service

Since the requests arrive over open connections, the TCP handshake cost is not included in this value. The values belong to a device serving only L7; if L4 vServices are also defined on the same device, HTTP requests/s reads about 15% lower.

SSL/TLS

Metric	Condition
TLS TPS (ECDSA-P256)	Full handshake, session ticket off, TLS 1.2, cipher suite ECDHE-ECDSA-AES128-GCM-SHA256

TLS operations are performed by the TR7 software by taking advantage of the hardware acceleration capabilities supported by the Intel platform. In each full handshake, the device generates one ECDSA P-256 signature and performs the key exchange again. The TLS TPS value achieved depends on the processor resources, the cryptographic algorithms used and the acceleration mechanisms active for these operations.

Throughput

Metric	Path
Throughput	Full proxy, 64 KB object
TLS throughput	TR7's own response, 64 KB, TLS 1.3 AES-GCM

Model	Throughput	Delivered to the client	TLS throughput
H17080	80 Gbps	40 Gbps	80 Gbps
H17120	120 Gbps	60 Gbps	120 Gbps
H17140	140 Gbps	70 Gbps	120 Gbps
H17160	160 Gbps	80 Gbps	140 Gbps
H17200	200 Gbps	100 Gbps	160 Gbps

Throughput is the sum of the two directions of the client and Backend Service legs; what is delivered to the client is half of this. The values are the traffic the processors can carry; the achieved throughput is also limited by the line rate of the installed network modules and by the PCIe lanes the modules are attached to. Reaching the processor capacity requires 100 GbE modules distributed across the two processors. The values are for 1500-byte frames. TLS throughput is the bytes sent by the network card and includes the TLS, TCP/IP and frame headers.

Software Compression

Metric	Condition
Compression throughput	gzip, pre-compression (input) bytes

The device compresses the response in gzip format, at the fastest level. The value depends on the repetition ratio of the content: highly repetitive content such as table and report output compresses noticeably faster than an ordinary HTML page.

Content class	Ratio	Compressibility
Heavily repetitive: table and report output	35.10:1	97%
Moderately repetitive: JSON and log	9.62:1	90%
HTML: repeating template	5.47:1	82%
Real report page	3.60:1	72%

L4

Metric	Condition
L4 HTTP RPS	HTTP requests forwarded over established persistent connections; 1 million persistent connections, 128-byte requests, 8 Backend Services
L4 Concurrent Connections	Connections opened and kept idle; the value is the number of entries in the device's connection table

In an L4 vService, the device does not terminate the connection and does not read HTTP content; it forwards the packets of the established connection to the Backend Service. In L4 HTTP RPS, persistent connections are opened at the start of the run and reused throughout the run; the value is the requests sent per second.

In L4 Concurrent Connections, connections are opened and kept open with a 60-second idle period; the value is not the generator's count but the number of entries in the device's own connection table. The measurement setup is limited to 100.7 M connections by the address space of the load generator; values above this limit are in the Capacity class.