Transcript IP Platforms Best Practices for Performance

IP Platforms Best
Practices for Performance
Pierre Lamy
Technical Lead Ottawa TAC
April 2010
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
Intro and revision history
This document describes methods and
techniques that users can apply on various
Check Point IP Security Appliances to achieve
optimal performance.
 Version 1.0 October 2009 Word format
 Version 1.1 January 2010 Word + PPT, minor
revisions
 Version 1.2 April 2010 Updates
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
2
2
General Performance Best Practices

These guidelines are Appliance independent and do not require any special
tuning.

Always use the latest versions of Check Point products. Always upgrade to the
most recent HFA (HotFix Accumulator) for a given version.

Create a small block of rules near the top of your rulebase, containing the
most heavily used rules. These rules should be fully accelerated with
SecureXL.

Keep the rulebase simple and small. Reduce the number of rules by
combining similar rules together. Rules which disable SecureXL acceleration
should be placed very low in the rulebase.

If not using VPN (encryption) on the module, make sure the VPN product is
disabled on that module.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
3
3
General Performance Best Practices

Do not use QoS from IPSO or Floodgate.

Avoid using Domain Objects. DNS lookup takes additional CPU cycles.

Avoid using UFP URI Filtering Protocol as this is resource intensive.

Use Networks instead of address ranges for Network Address Translation.

Keep logging to a minimum. Only business critical rules which will be
analyzed should have logging enabled. Drop rules, Accept rules, Stealth rules,
Cleanup rules and Implied rules should not log unless there is a clear
business case and the customer intends on analyzing the logs on a regular
basis. Otherwise logging should only be used for debug purposes.

IP Cluster members should have exactly the same package lists, having
dissimilar packages can cause state sync issues resulting in a performance
reduction.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
4
4
General Performance Best Practices

When you install an ADP interface module in an appliance, the network
processor in the card performs all VPN encryption and decryption, even for
VPN packets that ingress or egress through non-ADP interfaces. The built-in
Nokia encryption accelerator continues to accelerate IKE traffic but does not
perform any other processing.

If VPN traffic ingresses or egresses through a non-ADP interface, throughput
is negatively affected because the packets must transit the appliance’s
backplane to reach the network processor in the ADP module. It is
recommended that one configure VPNs to use only ADP interfaces to avoid
this performance loss.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
5
5
General Performance Best Practices

Uniprocessor systems (IP152, IP292, IP395, IP565, IP1265) should
use IPSO 4.2 (this is correct in April 2010) while Multiprocessor
systems (IP695, IP1285, IP2455) should use IPSO 6.2. The latest
build of any major release should always be used.

Multiprocessor systems should have Check Point R70 installed to take
full advantage of CoreXL technology. sk40465 has some more details
about this.

Use R70 + IPSO 6.2 on uniprocessor systems where there is a need
for a specific feature

Do not persist in the use of IPSO 4.2 or R65 once support is no longer
offered for these products.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
6
6
General Performance Best Practices

Using interface flow control can reduce network
throughput on busy interfaces and we do not suggest it be
enabled.

Avoid using SmartView Monitor to constantly monitor
system performance or collect historical data, as
SmartView Monitor itself has an impact on performance.

Avoid using custom scripts on systems which have
performance issues, as the scripts will incur CPU
resources.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
7
7
General Performance Best Practices

A CST from the IPSO system as well as a cpinfo from the
management station, are CRITICAL to provide to Check Point Support
when opening a case for assistance in troubleshooting performance
issues. Without at least those files, Check Point Support will be unable
to assist the customer. For systems with extremely high CPU, or
where running CST may cause problems, it is recommended to run it
with the following syntax: “nice +20 cst”. The command may take
hours to complete but will not divert critical system resources from
processing traffic in a live environment.

Any cpinfo that is provided to Check Point Support MUST be
generated using the latest cpinfo tool downloaded from the Support
site. This requires uninstalling the old cpinfo, and then installing the
latest one. Providing old cpinfo output to Check Point Support will
delay Support response.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
8
8
General Performance Best Practices

Avoid using Standalone installations. Separate the management station from
the enforcement point by running the management station on another system.

Use the default settings in the capacity optimization tab of the enforcement
point’s properties, changing only the total connections number.

General recommendations for these platforms are to use onboard quad ports
for:
 Security Gateway State Synchronization traffic
 Cluster protocol network traffic
 Policy and Appliance management traffic
 A path from the enforcement point to the Check Point Log server

SecureXL options should be matched between SXL and the Security Gateway
settings; for example Sequence Validation and Delayed Notifications.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
|
9
9
System Resources

High performance relies on the availability of
key system resources: CPU, memory, and
network interface bandwidth.

Tuning involves better using the current
hardware, not simply upgrading.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 10
10
System Resources - CPU

Small packet size traffic: The amount of traffic any network device can process
is not determined by byte throughput numbers, but rather by the packets per
second. A small packet uses as many resources as a large full size ethernet
packet.

CPU utilization is incurred on a per-packet level, rather than per-byte.
Therefore it is critical to note that a system that is processing a large number
of small packets, works as hard as a system processing the same number of
large packets. There may be a very large difference in the apparent byte
throughput between the systems.

State synchronization particularly demands high CPU cycles because the
nodes in a cluster perform synchronization for every connection they
encounter. This ensures high availability but causes high CPU usage. This is
especially important to consider, when deciding whether or not to synchronize
short lived connection types like DNS when using a VRRP pair – the failover
time of VRRP exceeds the DNS timeout. It would not be advisable to sync
DNS connections in that scenario.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 11
11
System Resources - CPU

High number of logging rules affects CPU. Logging uses CPU
cycles and is discouraged where there is no need.

The Active Log feature in SmartView Tracker will severely
compromise the ability of an enforcement point to process
traffic, and for performance reasons should not be used.

Accounting Logging: By default, accounting logging produces
two kinds of log tracking (one in Log Viewer and one in Account
Log View) for the same connection. Alerts as a tracking option,
also have a significant negative impact on performance.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 12
12
System Resources - CPU

Any configuration which disables SecureXL or forces traffic to
use Slowpath affects CPU. This includes rule configurations,
SmartDefense protections, and Floodgate.

NAT traffic incurs slightly more CPU impact that non-NATed
traffic. NAT in an ADP environment is strongly discouraged as
connection establishment rate acceleration does not work on
NAT traffic.

Traffic which is not connection-rate accelerated, uses more
CPU resources than traffic which is. Connection establishment
and teardown uses (relatively) a lot of CPU resources, even
when all SecureXL acceleration is in use.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 13
13
System Resources – Memory

Check Point recommends upgrading any Appliance’s memory to its full capacity to
improve performance.

The main factors that demand high RAM usage are:
 Concurrent connections
 Concurrent VPN tunnels
 NAT connections
 Security Servers

Use the Web User Interface to determine how much memory the Appliance has
installed. In the Web User Interface navigation tree, select Monitor --> System Utilization
--> CPU-Memory Live Utilization. Look for the “Total Real Memory” value. The top
command can also be used on recent IPSO versions to view the information on the
command line.

The amount of memory, allocated to the Check Point Security Gateway to process
network traffic, is determined under the capacity optimization tab of the gateway object
properties. The Automatic values should always be used, and the manual setting of
memory allocation should NEVER be used unless directed to by Check Point Support.

You should always set the capacity optimization to the maximum values (connections)
supported by the platform and memory configuration as detailed in the release notes for
IPSO. There is no drawback to doing this, but leaving the low defaults in place will
results in insufficient memory on busy systems.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 14
14
System Resources – Memory

Two factors can reduce the number of Security Gateway
connections that can be supported

Concurrent VPN tunnels are dependent upon the amount of
memory available in the Appliance. As you add more VPN
tunnels, the number of Security Gateway connections an
Appliance can support will decrease.

Security Servers will reduce the maximum number of
supported connections as they write to temporary files and
use 8 entries in the connections table.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 15
15
System Resources - Network Interface
Bandwidth

When the Security Gateway performance reaches the limitation of the
interface bandwidth and the CPU is still not fully utilized then the bottleneck is
the interface. One option to increase performance in this case is to use more
ports via Link Aggregation to achieve the maximum performance.

The limitation of the network interface is determined by the amount of packets
per second it can process. Assuming 1518 byte frames, a 100megabit NIC
port can sustain ~8234 packets per second (pps ) in each direction (full
duplex). A 1gb NIC port can sustain ~82340 pps in each direction (full duplex).

While the network port may sustain much higher numbers of pps than these –
and this is often seen in the field and QA – there is no guarantee that it WILL
support more than these standard pps numbers for a given link speed.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 16
16
System Resources - Network Interface
Bandwidth

If the number of packets per second exceeds those numbers, using Link Aggregation to
combine 2 or more links together is possible, thus increasing the amount of bandwidth
in pps that can be processed on that logical network interface.

IPSO Sync should not need more than a 1gb interface, and there can be problems
when IPSO Sync is run over a Link Aggregation group. The speed of the IPSO and
Check Point sync interface should be as fast as the fastest NIC port on the system. In a
10gb environment, typically 1gb for sync is sufficient. State sync should always be on an
isolated VLAN or network segment. Note that it is NOT supported to run Check Point
state sync over any interface VLAN other than 0 / 1 / untagged. Any VLAN’ing done on
the switch access port is fine, but trunking is not supported as Check Point state sync is
not designed to support the extra frame sizes.

Note: Security Gateway state sync link aggregated interfaces should be directly
connected using cross-over cables, unless you are using 3 or more cluster members.
IPSO and Check Point state sync should always be on an isolated VLAN or network
segment.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 17
17
IP Clustering

IP Clustering provides both high availability and scalability. IP
Clustering is useful when the performance of one system alone
is insufficient to provide the desired level of performance. For
example, when an Appliance CPU reaches ~30%, it would be
recommended to add another Appliance to form a two-member
cluster that can scale the Security Gateway performance. This
is a capacity planning exercise that Check Point Sales
engineers can help with. The 30% number is considered an
industry standard measurement or indicator that suggests more
capacity should be added.

IP Clustering is especially beneficial when using SmartDefense
features. With all SmartDefense features enabled, a twomember cluster’s HTTP transaction rate is about 40% higher
than a Standalone Appliance.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 18
18
IP Clustering

Use dedicated interfaces for cluster protocol networks and state
synchronization; do not share interfaces with the production traffic.

It is strongly recommended to use separate interfaces for cluster
protocol network and Security Gateway synchronization traffic so that
they are separate Broadcast domains.

Use a bandwidth of at least 100 Mbps full duplex for IPSO sync
interface(s). 1gb is recommended.

Use switches, not hubs, and never use crossover cables for IP
Clustering protocol networks.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 19
19
IP Clustering

Do not use IP Clustering Forwarding mode, when performance
is a concern. Unicast and Multicast provide better performance
and less latency. Forwarding mode is a fallback mode, for when
feature-poor network switches are in use.

If IGMP snooping is in use on the switch, disable it or configure
static CAMs in order to allow Multicast traffic on specific ports.

Use dynamic cluster work assignment for optimum load
balancing. This allows the cluster to periodically rebalance the
load by moving active connections between nodes.

Use delayed synchronization if your system processes many
short-lived connections, you are in VRRP or Standalone, and
SXL templates are in use. A 30 second delay in synchronizing
connections can boost the performance by about 20%. If you
use Check Point delayed notifications, you must also enable
SecureXL delayed notifications.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 20
20
ADP

Addition of ADP will increase performance of the appliance with some
limitations explained below. The decision to purchase ADP add on
cards should be made in consultation with Check Point Sales. ADP
should be considered if the performance improvement desired falls in
to one of the following categories:







Packet throughput performance, specifically for small packets.
Performance improvements for packet streams with mixed packet
sizes.
Encrypted traffic (VPN) forwarding.
Long-lived connections performance.
► For example data transfer rates for protocols like ftp, http etc.
NAT Performance only for long-lived NAT connections. (ADP
accelerates NAT throughput. Connection-rate Acceleration is
not currently supported for NAT connections. XMC cards
should be used instead for high NAT & CPS.)
Latency for both un-encrypted and encrypted traffic.
Multicast throughput performance.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 21
21
ADP

Performance issues with mixing ADP with non-ADP interfaces:

The best performance one can get is by not mixing traffic
between ADP and non-ADP interfaces. Running in mixed mode
will have performance impacts. When run in dual-mode –
having separate ADP traffic flows and separate non-ADP flows
one can see the Appliance performance scale as opposed to
not using any ADP interfaces at all.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 22
22
ADP Benefits

Throughput Acceleration:

The first packet in a connection is sent up the stack to the Security Gateway that
validated the packet based on the defined rule base. Once validated the Security
Gateway application tells IPSO via the SecureXL API to handle future packets in the
same connection. IPSO then instructs the ADP sub-system to create a bi-directional
flow for that connection. All future packets for that connection will now be processed by
the ADP sub-system.

The following protocols benefit from SecureXL & ADP Throughput acceleration:
 TCP, UDP, & traffic carried over those protocols
 IPSec VPN acceleration
 Multicast forwarding
 PIM (from IPSO 3.9 for IP2250 & IP2255; from IPSO 4.2 for all platforms)
 GRE & ESP
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 23
23
ADP Benefits

Connection Rate Acceleration

The first packet in a connection is validated by the Security Gateway application. Once
validated, the Security Gateway instructs IPSO to create a template so that IPSO can
validate future connections where only the Source Port differs. A template consists of
the following attributes: SrcAddr, SrcPort, Proto, DestAddr, & Dest-Port. IPSO compares
the first packet in the next connection to its template table. If the packet matches a
template then IPSO adds the connection to its table, then instructs ADP to create a bidirectional flow for the connection and lastly informs the Security Gateway about the
new connection. All future packets are processed by the ADP module.

The following protocols benefit from SecureXL & ADP Connection-rate acceleration:
 Unencrypted TCP, UDP, & traffic carried over those protocols
 Particularly effective on HTTP 1.1 traffic
 Even more effective on HTTP 1.0 traffic
 HTTP 1.0: Separate connection for each HTTP component
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 24
24
ADP Best Practices

Configure traffic to flow in/out of the same ADP subsystem, since
traversing to another ADP subsystem or worse to a non-ADP interface
will negatively impact throughput performance. There is a 10gb full
duplex bandwidth over the crossbar between 2 ADP subsystems.

Do not use ports connected to the ADP subsystems for cluster
protocol network or for Security Gateway state synchronization. Use
onboard ports for Security Gateway state synchronization; this will
guarantee that the synchronization data goes to its own channel and
will avoid sync packets being lost. This also prevents the sync data
from disrupting the data passing between the ADPs and the main
CPU. Likewise do not log to the management station via the ADP
interfaces.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 25
25
ADP Best Practices

Note that the backplane connecting the ADP subsystems to the
main CPU has limited bandwidth, and this bottleneck will impact
throughput performance when there is a lot of non-accelerated
traffic.

Do not combine non-ADP ports and ADP ports in a link aggregation
group or redundancy group.

Do not include interfaces on different ADP I/O cards in the same
link aggregation or redundancy group. IP Security Appliances do
not support cross ADP link aggregation.

SecureXL is enabled by default. Note that it’s critical not to disable
SecureXL because SecureXL is required for ADPs to function.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 26
26
ADP Best Practices

Avoid performing tcpdump or fw monitor on ADP
interfaces when the interfaces are under heavy load.
Performing a tcpdump or fw monitor on an ADP interface
forces all traffic received or transmitted by the ADP
subsystem, to be copied and piped to IPSO through the
backplane. SecureXL will still be used,and the ADP will
still accelerate, but the backplane will be choked with
data. This causes a significant degradation in
performance due to constricted backplane capacity.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 27
27
Limitations of ADP

Traffic that is not throughput or connection rate
accelerated will not benefit from ADP acceleration. All
limitations of SecureXL apply to ADP.

Transparent mode will accelerate traffic normally, however
a special design consideration is that there must be routes
pointing out of the xmode interfaces, as SecureXL
depends on caching route table lookups.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 28
28
Limitations of ADP

Enabling Sequence verifier: this solution requires enabling sequence verifier option on
IPSO as well as in SmartDashboard.

This solution was suggested after analyzing the CST, where it was observed that most
(60 million out of 80 million) of the TCP connections were getting closed with RSTs
instead of the usual 3-way handshake for terminating TCP connections. As part of the
stateful inspection, the Security Gateway needs to monitor all TCP RSTs if sequence
verifier is not turned ON, as they are categorized as untrusted RSTs. This behavior of
terminating TCP connections causes additional load on ADP backplane interfaces,
where the packet drops were observed.

By turning ON sequence verifier ADP will perform sequence verification on all TCP
connections thereby validating even RSTs that are used to terminate the TCP
connections. Once the TCP RST is validated by ADP and is accepted there is no need
to send this packet to the Security Gateway, thereby reducing the backplane traffic and
the overhead of the Security Gateway having to inspect these packets.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 29
29
Limitations of ADP

Once the sequence verification is turned ON, you should see significant
reduction in packets going over the backplane to the Security Gateway. This
can be monitored by executing the following command


ipsctl –i net:dev:adp:if:stats with an “r” option to monitor the rate of tcp_rst
As a result of the reduction in TCP RSTs going over the backplane, we should
observe less drops of data packets on backplane which can be monitored by
executing the following command.

ipsctl –i net:dev:bp:msg:stats with an “r” option to monitor the rate of
rx_fc_drops
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 30
30
Limitations of ADP

Too many control messages queued up on eth1, will result in data loss on the
data channels eth2-4. The queue depth for the control channel is tunable, the
default is 64 in IPSO 6.2. Potential values are 128 and 256:


You should see a decrease in the rate of dropped packets on data channel.
This can be monitored by executing the following command:


ipsctl -w net:dev:bp:msg:delay_drop_limit 128
ipsctl –i net:dev:bp:msg:stats with an “r” option to monitor the rate of
rx_fc_drops
If this solution does not yield the desired result then the delay_drop_limit can
be easily set back to its default value of 64. Setting of these ipsctl variables
takes effect immediately and is non-intrusive.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 31
31
Limitations of ADP

Turning off the delay_drop variable: this solution requires changing the default value of
delay_drop, an ipsctl tunable in IPSO. IPSO pro-actively drops data packets, when control
channel is congested. This option can be completely turned off by executing the following
command

ipsctl –w net:dev:bp:msg:delay_drop 0

The option of dropping data packets when control channel is congested was developed under
certain performance benchmarking conditions, where the box is tested for limits and the
aggressive load conditions persist for an extended period of time. This is the reason for
aggressive drop_delay_limit to 64 by default.

Unfortunately, this condition also comes into effect when there is transient congestion on the
control channel. By turning off this feature, we do not drop the data packets pre-maturely.

Current congestion level and the max congestion level on the control channel can be
monitored by executing the following command:


ipsctl –i net:dev:bp:msg:stats with an “r” option to monitor the rate of bms_scheds and
bms_scheds_max
If this solution does not yield the expected result, then you can revert back to default behavior
immediately by executing the following command.

ipsctl –w net:dev:bp:msg:delay_drop 1
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 32
32
Limitations of ADP

The PSL acceleration feature should be enabled on multicore
systems using SecureXL and CoreXL, with or without ADP. The
ipsctl tunable can be found using
 ipsctl –a net:sxl

PSL acceleration allows full acceleration of all but the last
packet containing the application level Protocol Data Unit
IPS / SmartDefense takes care of a go/no-go to drop denied
connections.

©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 33
33
Security Gateway Performance Tuning

NAT

IP Appliances do not support Network Address Translation
(NAT) connection acceleration. The first packets of the first
connection on the same service are forwarded to the Security
Gateway application. Then a “template” of that connection is
created so that subsequent TCP establishments on the same
service, where only the source port is different, will be
accelerated by SecureXL. NAT connections setup and teardown
cannot be accelerated because NAT templates are not
supported.

While each connection uses two entries in the flows table,
connections involving NAT use four entries. NAT connections
use more CPU and memory resources compared to ordinary
connections.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 34
34
Security Gateway Performance Tuning

Rulebase Size

Although there is no limit to the number of rules in a Security Gateway database, there
is a performance impact as the number of rules grows. The more rules an Appliance
has, the more it will cost the Appliance in compilation time and runtime efficiency.

Rulebase size affects connections rate performance.

Rulebase order is important and can affect performance. Use the following guidelines
for organizing the rulebase:
 The rulebase should be as simple as possible. With fewer rules the rulebase will be
more efficient and less error prone.
 When creating a rule, be specific. Narrow down the source, destination, and
service.
 Avoid using “Any” in the service field.
 The most active NAT rules should be at the top of the NAT rulebase.
 Defining Group Objects for networks allows the policy compiler to superset traffic in
the actual rule for a performance gain.
 Anti spoofing should be configured for all the Security Gateway interfaces.

Avoid using “negate” in the rulebase (For example, a network exclusion)
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 35
35
Performance Troubleshooting

Follow the guidelines to troubleshoot performance issues and use the
best practices outlined above to optimize the Appliance overall
performance.

Do NOT use fw monitor for performance troubleshooting. Connecting a
span port via a switch, is preferred to tcpdump. Traffic captures sent to
Check Point Support should not exceed 80mb uncompressed. Particular
lines within a packet capture should be referenced by the customer as
needed.

Check free disk space using the “df –k” command line tool.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 36
36
Performance Troubleshooting

Check currently used CPU statistics using the “vmstat 1” tool. The last
3 columns are significant, customers should never concern
themselves with the other columns. The very last column is CPU Idle
time, this is the amount of free CPU cycles, in percentage, since the
last vmstat iteration. The second-last column is System CPU usage in
percentage, this includes IPSO and Check Point kernels as well as
interrupts. The third-from-last column is User CPU utilization, this is
usually due to Policy Installation, SmartDefense, Security Servers, and
user scripts or commands.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 37
37
Performance Troubleshooting


The “top” command line utility is used on recent IPSO versions, this
utility will provide per-process CPU and memory utilization, as well as
global statistics, and more granular CPU statistics such as % of
interrupts. The percentage of interrupts includes both software
interrupts and hardware interrupts. Hardware interrupts are virtually
never the cause of performance issues, performance issues are
virtually always caused by software interrupts.
If it can be shown that the performance problem is due to a high ratio
of interrupts compared to overall CPU utilization, the fix is always first
to properly tune SecureXL.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 38
38
Performance Troubleshooting

SecureXL Acceleration statistics can be verified using the following
commands:





fwaccel
fwaccel
fwaccel
fwaccel
stat
stats
stats –s
templates –s
These commands should provide a good overview of how much SecureXL is
in use. The SecureXL and Nokia IPSO Guide
(http://downloads.checkpoint.com/dc/download.htm?ID=10036) should be
used to help tune the rulebase, and ensure that as many connections and
packets are accelerated as possible.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 39
39
Performance Troubleshooting


If dropped packets are a concern, they can be checked as a snapshot in IPSO
by running the command “ipsctl –a ifphys | grep qdrop”. This will provide an
indication of which interfaces are dropping traffic. Please note that these
counters are incremental since system boot time. For more information about
qdrops, consult sk39462.
To view realtime statistics for a particular counter, run ipsctl –i <full-counterpath> use use the “r” command to toggle the rate per second counter refresh.

Any qdrop which is logged will have a corresponding reason code
incremented, in
 ifphys:<interface_name>:errors
 ifphys:<interface_name>:stats

Common drops are rx_mpc, which represents the Receive, Missed Packet
Count, and symerrs, Symbol Errors.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 40
40
Performance Troubleshooting

rx_mpc is due to the operating system not being able to flush the receive
buffer for the interface fast enough. The receive buffer queues up incoming
data which has been received on the physical media, and is flushed from the
queue every time there is an interrupt. The interrupt is triggered under 2
conditions – when the rx_ring has reached ¼ full, or after a timer is hit. More
information about these processed is detailed in sk39176. There are advanced
tunable variables that can be used under direction of Check Point Support to
influence this behavior.

Symbol Errors are due to a bad fiber network cable, dirty or dusty NIC port or
fiber connectors. It can also be due to a bad NIC card or switch port but this is
very uncommon. Symbol errors only increment for received data for the local
side, not sent data. More information about symerrs is detailed in sk39733.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 41
41
Performance Troubleshooting


A lot of short lived connections transiting the enforcement point can
cause slowdowns, as connection establishment and teardown incurs
CPU utilization. This can be partly mitigated, by ensuring templates
can be created for the most heavily used rules. You may also be able
to use the Fast Expire SecureXL feature.
Fast Expire should be used primarily for short lived connections such
as DNS.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 42
42
Performance Troubleshooting

Use ifconfig to verify that no interfaces have the PROMISC flag set. An
interface in promiscuous mode forwards all frames seen on the physical
media, to the operating system for Layer 2 filtering. An interface in nonpromiscuous mode, uses a MAC chip to filter Layer 2 frames to ensure that
only frames which are destined for the local machine are passed to the
operating system. This is determined based on Unicast/Broadcast/Multicast
MAC Address lists in the Receive Address High and Low registers. PROMISC
is set for Transparent Mode and this is normal behavior.

Customers may wish to view the Security Gateway connection tables in
human readable format to help with rules optimization. Check Point Support
has internal-only tools to read the output of “fw tab –u”. Customers may be
interested in an unsupported third-party script which can be found at
http://www.fw-1.de/aerasec/download/fw1-tool/fw1-tool.pl

Check Point makes no guarantees about this product, and provides this
information as reference only.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 43
43
Advanced debugging and tuning

Advanced debugging and tuning should only
be carried out under the direction of the
Escalations group, or Development.
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 44
44
The End – Questions?
©2010 Check Point Software Technologies Ltd.
|
[Unrestricted] For everyone
| 45
45