These notes are not written to educate on different technologies but to summarize best practices from the SRND.
1. General Notes
a. Qos CPU load should never exceed 75 %
b. RTP should be equal to or les then 33% of link bandwidth
c. RTP on-way Jitter should not exceed 30ms of delay
d. LFI (MLP/FrF.12) should be applied to all links less than or equal to 768Kbps
e. cRTP should be applied to all links less than or equal to 768Kbps
f. Default policy map only allows for 75% of bandwidth to be allocated.
g. IGP don’t show under queue statistics because of PAK_priority but EGP do because they don’t fall under PAK_priority mechanism
h. L2 bandwidth = [(Packets per second] * (X bytes for voice payload + 40 bytes for RTP/UDP/IP headers + Y bytes for L2 overhead) * 8 bits] / 8
i. L3 bandwidth = [(Packets per second] * (X bytes for voice payload + 40 bytes for RTP/UDP/IP headers) * 8 bits] / 8
j. Packets per second = [1/(sampling rate in msec)] * 1000
k. Voice payload in bytes = [(codec bit rate in kbps) * (sampling rate in msec)] / 8
l. 2L Headers (QOS SRND)
i. Ethernet = 32 bytes
ii. Frame-Relay = 4bytes / (FRF.12 = 8 bytes)
iii. MLP = 13
m. Voice Codec Accounting
i. G.711 - GK = 128 Region=80 RSVP=96
ii. G.729 - GK =16 Region=24 RSVP=40
n. Signaling Accounting ( 25% over provisioning) (UCM_SRND)
i. SCCP / GW = Bandwidth (bps) = 256 * (# of IP phones and GWY’s)
1. Encrypted Signaling = Bandwidth (bps) = 415 * (# of IP phones )
ii. SIP = Bandwidth (bps) = 538 * ( # of IP phones and GWY’s)
1. Encrypted Signaling = Bandwidth (bps) = 619 * (# of IP phones )
iii. Smallest amount that can be configured on IOS GW is 8kbps per queue.
2. Wan Notes
a. Voice and Data Models
i. Three-Class
1. Voice
a. Set at EF with a maximum of 33 % of the link bandwidth incorporating CaC with L2 calculations.
2. Call-Signaling
a. Set at CS3/AF31 and is set a generic 5% of remaining bandwidth or at specific CaC needed signaling rates.
3. Best-Effort (class-default)
a. Uses Fair-queuing and is set at 62% of remaining bandwidth
ii. Five-Class
1. Voice
2. Call-Signaling
3. Critical Data
a. Incorporates Transactional Data(AF21) and IGP/Network Management(CS6/CS2) at 36 % of link with WRED enabled
4. Scavenger Traffic
a. Marked at CS1 and is policed down to 1% of link.
*If class-default does not have bandwidth statement Scavenger Traffic queue still has the ability to steal bandwidth from it. And on older non Distributed platforms bandwidth preludes fair-queue and enables FIFO. This is not true on newer platforms.
5. Best-Effort
iii. Five + Class
1. Dual LLQ (Voice , Interactive-Video AF41)
b. Link Speed
i. Slow ( link speed < 768 kbps) : (MLP LFI / FRF.12 and cRTP)
1. General Slow Speed Notes
a. Deployment of Interactive-Video generally is not recommended on these links because of serialization implications.
b. These links require LFI to be enabled if VoIP is to be deployed over them.
c. cRTP is recommended (with a watchful eye of CPU levels.)
d. Check Tx-ring sizes ( especially on slow-speed PVCs); tune to 3, if needed.
e. Three- to five-class traffic models are recommended
2. Leased Lines
a. Fragement delay command will limit delay
3. Frame-Relay ( FRF.12 is required, cRTP optional)
a. Maximum of 10ms serlization delay
i. Fragment Size = PVC_kbps * Max Jitter_ms / 8
1. 70 Bytes = 56kbps * 10ms / 8
b. Voice-Adaptive Traffice Shaping ( VATS) optional (UCM_SRND)
c. Voice-Adaptive Fragmentation (VAF) optional(UCM_SRND)
ii. Medium ( 768 kbps < link speed < T1/E1 ):
1. General Medium Speed Notes
a. Voip or Interactive-Video can be assigned to LLQ (usually, there is not enough bandwidth to do both and still keep the LLQ provisioned at less than 33 percent- alternatively, Interactive-Video can be placed in a CBWFQ queue).
b. LFI is no required
c. Three- to Five-class traffic model are recommended
2. Leased Lines ( MLP, HDLC, PPP and cRTP optional)
a. Bandwidth Statements for bundled MLP should be under Individual interfaces instead of Multi-Link interface for correct calculations.
3. Frame-Relay (FRF.12 is not required , cRTP optional)
iii. High ( T1/E1 link speeds):
1. General High Speed Notes
a. LFI is not required
b. cRTP generally is not recommended ( because the cost of the increased CPU levels.)
c. Five to 11 class traffic models are recommended
2. Leased Lines
a. Use MLP bundling ( CEF per-destination/packet, MLPP bundles)
3. Frame-Relay
a. Use IP CEF per-packet load balancing
c. Frame-Relay
i. Committed Information Rate
1. Set the CIR to 95% of PVC to contracted rate because Contract Rates do not take into account FR overhead/CRC.
ii. Committed Burst Rate
1. Set the Bc to CIR/100 on non-distributed and CIR/125 on distributed platforms.
a. BC = CIR/8
2. Tc ( time traffic is actually sent) Tc = Bc / CIR
a. 10ms = 5600 / 56000
i. TC = 10ms
ii. Bc = 5600 ( configured )
iii. CIR = 56000Kbps
iii. Excess Burst Rate
1. Set the Be to 0.
a. Good in closed networks but not for Carrier based because receiving end doesn’t get Bc+Be it only receives Bc.
iv. Minimum Committed Information Rate
1. Set minCIR to CIR.
a. minCIR comes into play when BECN’s are received and by default CIR is set to CIR/2. This is not desired for voice and should be disabled unless voice-adaptive traffic shaping is deployed.
v. Slow ( link speed < 768 kbps) : ( FRF.12 )
1. Maximum of 10ms serialization delay
a. Fragment Size = PVC_kbps * Max Jitter_ms / 8
i. 70 Bytes = 56kbps * 10ms / 8
d. UCM Cluster
i. Maximum delay should not exceed 40ms one-way delay or 80ms round trip inter-cluster.
Reference
No comments:
Post a Comment