| max. bandwidth | max routers connected | Technology | Functionality | Classification & marking | Architecture | Management and ease of use | Monitoring | ease of use | Security | Fault tolerance | High-availability | |
| Cisco | platform dependent | per-router behavior | Queuing. Packets are placed in holding queues and retransmitted.
The retrasmission is based on CBQ and WFQ algorithms. When traffic is overflowing the queus, packets are discarded to reduce network congestion. Cisco IOS is using mainly 2 queuing techniqes: WFQ – setting percentage of an interface's available bandwidth to be used by different typesof traffic. Priority Output Queuing – sending traffic out on an interface based on priority. One-way stream control on per-hop basis - incoming traffic has to be managed at the remote sites. |
Shaping and policing were tested and worked as expected. Unable to supply shaping policy within another shaping policy. Bandwidth reservation ("bandwidth" service-policy inside a "shape" service-policy) work as expected. LLQ works fine. |
Layers 3, 4, 7 (NBAR). Layer 7 is very CPU intensive. By protocol family, protocol, layer 7 parameters, acl match, source, destination, qos-group | Can be applied on specific interface \ subinterface, thus
enables control on specific point.
Can control traffic between sites in star topology. |
Easy by QDM (router-based configuration software). A bit
complicated by QPM (overall policy configurator). QDM was unables to configure shaping policy on a 3662 router running IOS 12.2.16, while command-line enables this. QDM can "reverse engeneer" currently configured classes and policies. |
Provides monitring on applied policies only. No auto-discovery of traffic. Bandwidth utilization graph of 300sec average only - unable to provide current bps rates. Unable to differentiate beteen incoming and outgoing traffic rates. - provides outgoing rates only. |
very uneasy - deployment on large networks (20+ routers) is
usually made by QPM configurator. The current version (QPM 3.1) is very not
friendly, very hard to use and works very slow even on LAN because of
inherent SSL session between client and server. Deployment on small networks is made by CLI. In this case, no report tool is available, and one has to build reporting capabilities by himself. |
access-list, radius, local user database. | none | HSRP |
| packeteer | 1Gbps max | unlimited (can be connected directly to a router, or to a VLAN containing multiple routers). | Tcp rate-limit (end-to-end QoS policies). Interferes TCP flows and reducing transmission rates by changing TCP's window size. | Works as expected - can limit bandwith for a single\group of application. Can guerentee minimum bandwidth to a single\group of applications. Unused bandwidth is shared on demand. | Layers 3, 4, 7. classification by protocol family, protocol, port, host, host list, subnet. Can't classify by list of subnets. Marking is available in shaping mode only (Can't use PacketSeeker for this task). | Best controls TCP streams. Might be uneffective on short
streams, and increases ACK overhead in those cases. Can't control traffic between sites in star-topology, unless placing boxes at the edge. |
Very easy and intuitive either by HTTP to the box, or by central "policy center" software. | Supports auto-discovery, which has to be further evaluated -
some streams were classified as "dafault class" instead of a
specific detailed class. Provides monitoring on auto discovered and pre-defined classes in a very easy way, without the need to enable any policies. Provides historical graph\details of current\1min. average bps rates. Can differentiate between incoming and outgoing traffic rates. Details are displayed in a very clear and simple way. |
Very easy. Configuration is made by HTTP session to the box, GUI is very user friendly and reporting capabilities are very excellent. | access-list, radius, local user database. | becomes a "wire" when off\unplugged. can use LEM interface for same-box redundancy (using STP). |
none |