R8x - Gateway Performance Metrics

[HeikoAnkenbrand]

Intruduction

In the last weeks I have been asked again and again how I can increase the performance of my Check Point gateway. Now comes my counter-question. What do you want to reach in Performance Tuning?

Therefore, I have created an overview of what the goal is!

Chapter

Moe interesting articles:

- R80.x Architecture and Performance Tuning - Link Collection
- Article list (Heiko Ankenbrand)

Performance Metrics

In principle, there are several performance metrics:

• Throughput (Bandwidth)
• Connection rate
• Packet rate
• Concurrent connections
• Latency

There are standardized test procedures according to RFC for this:

	Throughput	Connection rate	Packet rate	Concurrent connections	Latency
RFC	RFC3511 5.1.4.1	RFC3511 5.3.1	RFC3511 5.1.4.1	RFC3511 5.2.4.2	RFC2544 26.2
Units	Bit/s	Connections/s	Packets/s	Absolute number of connections	(m)s
Testing conditions	Large UDP	Small TCP	Small UDP	Small TCP	Small UDP
Bottleneck	Bus, Interfaces	CPU	CPU	Memory	Bus, Interfaces, CPU, Infrastructure

Throughput

Description: RFC3511 – 5.1.4.1

Throughput: Maximum offered load, expressed in either bits per second or packets per second, at which no packet loss is detected. The bits to be counted are in the IP packet (header plus payload); other fields, such as link-layer headers and trailers, MUST NOT be included in the measurement.

Units: Bits per second

Testing conditions for achieving best results: Large UDP

Bottleneck: Bus, interfaces

Connection Rate

Description: RFC3511 – 5.3.1

To determine the maximum TCP connection establishment rate through or with the DUT/SUT, as defined by RFC 2647 [1]. This test is intended to find the maximum rate the DUT/SUT can update its connection table.

Units: Connections per second

Testing conditions for achieving best results: Small TCP (HTTP 64B)

Bottleneck: CPU

Packet Rate

Description: RFC3511 – 5.1.4.1

Throughput: Maximum offered load, expressed in either bits per second or packets per second, at which no packet loss is detected. The bits to be counted are in the IP packet (header plus payload); other fields, such as link-layer headers and trailers, MUST NOT be included in the measurement.

Units: Packets per second

Testing conditions for achieving best results: Small UDP

Bottleneck: CPU

Conncurent Connections

Description: RFC3511 – 5.2.4.2

Maximum concurrent connections: Total number of TCP connections open for the last successful iteration performed in the search algorithm.

Units: Absolute number (amount)

Testing conditions for achieving best results: Small TCP (HTTP 64B)

Bottleneck: Memory

Latency

Description: RFC2544 – 26.2

The latency is timestamp B minus timestamp A as per the relevant definition from RFC 1242, namely latency as defined for store and forward devices or latency as defined for bit forwarding devices.

Units: (m)seconds

Testing conditions for achieving best results: Small UDP

Bottleneck: Interfaces, Infrastructure, CPU, Bus

Analysis of metrics

The analysis of the above mentioned parameters is very easy with the command cpview.

# cpview

On 41K, 44K, 61K, 64K or Maestro systems use:

# asg perf -v

[Mehmet_Gul]

There are tools that you can use to generate traffic to test the performance parameters?

[HeikoAnkenbrand]

I use iPerf for throughput testing.

Link: 
iperf.fr 

[Julius_Dockward]

How can I use the iperv tool?

Are there any examples here?

[HeikoAnkenbrand · Accepted solution]

Use an opern server and a client.

# iperv3 -s     > iperv server

# iperv3 -c <iperv server ip> -n 64          > iperv client for small tcp packets 

# iperv3 -c <iperv server ip> -u -n 64           > iperv client for small udp packets 

# iperv3 -c <iperv server ip> -u -n 1460          > iperv client for large udp packets 

[Ralph_Heckler]

Great information.

[Hong_Ning]

Hi @HeikoAnkenbrand 

we use 4 servers (2xclient and 2xserver)  for performance tests.

But we get only a throughput at 10 GBit/s interface from 3 GBit/s on the firewall on a open server HP DL 380 G9.  What could be the problem?

[HeikoAnkenbrand]

3-4 GBit/s is normal. If you need more throughput, you should enable multi queueing in the first step.

[HeikoAnkenbrand]

More read here:
R80.x - Performance Tuning Tip - Multi Queue

[Maxi_Pertus]

Or enable more CoreXL instances!

[h_hong]

I cannot find the iperv3 tool on the gateway.

[_Val_]

Why do you need that on a GW in the first place? You use client to server connections through the GW to test performance.