Information here is by rumor, innuendo and extrapolation. Manufacturers rarely put info on packet buffers in their data sheets. There are some summary thoughts |
||||||
The buffer size question discussed in 2012 on the nanog list and is reproduced. Buffer requirements for long RTT networks is less well understood than you might hope. IETF tests for burst management are not as well developed as bandwidth tests. There is a draft that hints at progress. Packet pacing can improve buffer effectiveness by making TCP less bursty. | ||||||
Packet buffers and switch performance have shown steady improvement. This table attempts to show this for ASICs with integrated packet buffers and why using crufty old stuff is at your peril. |
||||||
Incast is a buffer exhaustion phenomena that is one consequence of running out of packet memory. |
||||||
Shared memory means that the hardware permits buffers to be used by any port that needs them. An intel white paper compares shared memory with other architectures. In a shared memory design it is not possible to let ALL the memory go to queued packets. There would be no room for new arrivals which would lead to head of line blocking. The other major option for a switch fabric is a crossbar matrix. |
||||||
Buffer queue depth monitoring cannot be done directly with SNMP MIB-II polling of the current occupancy of the buffer. Even if a buffer depth SNMP poll object existed it would not possible to interrogate it on a time scale short enough to catch microbursts. A burst that would fill a 4 MByte buffer would completely drain in 3.2 mS at 10 Gb/s. You could hope for indirect evidence of buffer exhaustion by monitoring packet drops. Bursts too short to cause drops can nonetheless be long enough to affect performance. Direct queue monitoring can thus add valuable information. |
||||||
Some switches have multiple switch ICs that each manage their own memory pool. Examples are the Brocade FCX648S and the Cisco 3750-48. Memory from one IC can be shared among the ports in that IC's group but cannot be loaned out to ports controlled by other switch chips. Here we are interested in queue resources that can be claimed by a single flow for burst absorption -- not the total RAM in the system. |
||||||
Tolly (tolly.com) occasionally reports on the ability of switches to sustain microbursts in his reports on data center switches. These measurements relate directly to output port buffering. See esp the IBM G8264 below. |
||||||
Max buffer queue depth requires that all packet memory can be put into a single queue. QoS schemes divide buffer resources among defined queues. As such, I am not interested in the QoS descriptions and these are even less reliable than the rest of this doc. |
||||||
Model |
Port Type |
RX Queue |
TX Queue |
Total Buffer |
RX Buffer |
TX Buffer |
Trident+ Shared Memory |
||||||
48 SFP+ and 4 QSFP+ |
8Q |
9 MB |
5? MB |
|||
48 SFP+ and 4 QSFP+ |
8Q |
9 MB |
||||
48 SFP+ and 4 QSFP |
8Q |
9 MB/switch |
5 MB |
|||
Dell 8132F & 4032F |
24 SFP+ and 2 x QSFP+ |
8Q |
9 MB |
|||
Dell 8164F & 4064F |
48 SFP+ and 4 x QSFP+ |
8Q |
9 MB |
|||
48 SFP+ and 4 x QSFP+ |
9 MB |
|||||
48 SFP+ and 4 x QSFP+ |
9 MB |
|||||
48 SFP+ and 4 QSFP+ |
12Q |
9 MB |
5 MB |
|||
48 10GTw-Pr and 4 QSFP+ |
8Q |
9 MB |
||||
48 SFP+ and 4 QSFP+ |
8Q |
9 MB |
||||
48 SFP+ and 4 QSFP+ |
8Q |
9 MB/switch |
||||
48 SFP+ and 4 QSFP+ |
4Q |
9 MB |
||||
Allied Telesis DC2552 |
48 SFP+ and 4 QSFP+ |
8Q |
9 MB |
|||
48 SFP+ and 4 QSFP+ |
||||||
48 SFP+ and 4 QSFP+ |
9 MB |
|||||
48 SFP+ and 4 QSFP+ |
9 MB |
|||||
48 1000-base-T and 4 SFP+ |
8Q |
9 MB but see notes |
5? MB to 1 port | |||
Brocade TOR custom |
||||||
48 SFP+ and 4 QSFP+ |
24 MB |
|||||
96 SFP+ and 12 QSFP+ |
24 MB |
|||||
varies by model |
12.2 MB |
|||||
48 SFP+ and 6 QSFP+ |
12.2 MB |
|||||
32 QSFP+ |
8Q |
12 MB |
Dynamic up to 8 MB |
|||
64 QSFP+ |
8Q |
12 MB/ASIC, 48 MB total |
Dynamic up to 8 MB |
|||
16 slot: 512 QSFP |
8Q |
12 MB/ASIC |
Dynamic up to 8 MB |
|||
26 QSFP+ and slot |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 6 QSFP+ and expansion |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 10GbaseT and 6 QSFP+ and expansion |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
12 MB |
8 MB |
|||
44 SFP+ and 6 QSFP+ |
8Q |
|||||
32 QSFP+ |
8Q |
No info |
No info |
|||
32 QSFP+ or 31 QSFP+ and 4 SFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
12 MB |
Dynamic up to 8 MB |
|||
32 QSFP+ |
8Q |
12 MB |
Dynamic up to 8 MB |
|||
24 QSFP+ + expansion slots |
8Q |
12 MB |
Dynamic up to 8 MB |
|||
24 SFP+ + 4 QSFP + expansion slots |
8Q |
12 MB |
Dynamic up to 8 MB |
|||
32 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 4 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
32 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
32 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
16 QSFP+ |
8Q |
12.2 MB |
Dynamic up to 8 MB |
|||
48 SFP+ and 4 QSFP28 100G and 2 QSFP+ |
8Q |
16 MB |
Dynamic up to 12 MB |
|||
48 SFP+ and 6 QSFP 40G |
8Q |
16 MB |
Dynamic up to 12 MB |
|||
Three switch configurations |
8Q |
|||||
8Q |
Dynamic up to 12 MB |
|||||
8Q |
Dynamic up to 12 MB |
|||||
8Q |
Dynamic up to 12 MB |
|||||
32 QSFP+ or 26 QSFP+ and 6 QSFP28 |
8Q |
Dynamic up to 12 MB |
||||
48 SFP+ and 6 QSFP28 |
8Q |
Dynamic up to 12 MB |
||||
40 1/10-Gbps SFP+ and 4 QSFP28 100G |
8Q |
16 MB |
Dynamic up to 12 MB |
|||
80 1/10-Gbps SFP+ and 4 QSFP28 100G |
8Q |
16 MB |
Dynamic up to 12 MB |
|||
32 40-Gbps QSFP+ |
8Q |
16 MB |
Dynamic up to 12 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
16 MB |
Dynamic up to 12 MB |
|||
48 SFP+ and 6 QSFP+ |
8Q |
16 MB?? |
Dynamic up to 12 MB |
|||
48 SFP+ and 4 QSFP28 100G |
8Q |
16 MB |
Dynamic up to 12 MB |
|||
2-port 100Gb/s module, 8-port QSFP module, . . . |
8Q |
|||||
32 port 100G QSFP28 and 2 port 10 Gb/s |
8Q |
32 MB |
27 MB shared pool |
|||
48 port 25G SFP28 and 12 port 100 Gb/s |
8Q |
32 MB |
27 MB shared pool |
|||
48 port 25G SFP28 and 8 port 100 Gb/s |
8Q |
32 MB |
||||
48 port 25G SFP28 and 8 port 100 Gb/s |
8Q |
32 MB |
||||
32 port 100G QSFP28 |
8Q |
32 MB |
||||
48 port 25G SFP28 and 8 port 100 Gb/s |
8Q |
|||||
48 port 25 Gb/s SFP28 and ??? 100G QSFP28 |
8Q |
32 MB |
||||
32 port 100 Gb/s QSFP28 |
8Q |
32 MB |
||||
48 port 10G QSFP and 6 port 100 Gb/s |
8Q |
options: 20-50 Mbyte |
||||
48 port 25G QSFP28 and 6 port 100 Gb/s |
8Q |
options: 20-50 Mbyte |
||||
32 port 100G QSFP28 |
8Q |
options: 20-50 Mbyte |
||||
64 port 100G QSFP28 |
8Q |
options: 20-50 Mbyte |
||||
48 10G-base-T and 4 QSFP28 and 2 QSFP |
8Q |
?? MB |
||||
48 SFP+ and 4 QSFP28 and 2 QSFP |
8Q |
?? MB |
||||
48 port SFP+ and 2 QSFP+ and 2 QSFP28 |
8Q |
12 MByte |
8 MB |
|||
48 port SFP+ and 4 QSFP28 |
8Q |
12 MByte |
8 MB |
|||
32 port QSFP28 100 Gb/s |
8Q |
16 MByte |
Est 3 MByte per core |
|||
32 port QSFP28 100 Gb/s |
8Q |
16 MByte |
Est 3 MByte per core |
|||
Depends on modules selected. |
8Q |
16 MByte |
Est 3 MByte per core |
|||
48 port SFP28+ 25 Gb/s + 6 x 100G |
8Q |
16 MByte |
Est 3 MByte per core |
|||
32 port QSFP28 100 Gb/s |
8Q |
16 MByte |
Est 3 MByte per core |
|||
32 QSFP28 100G |
8Q |
16 MByte |
Est 3 Mbyte per core |
|||
32 port QSFP28 100 Gb/s and 2 SFP+ |
8Q |
16 MByte |
Est 3 MByte per core |
|||
64 port QSFP28 100 Gb/s and 2 SFP+ |
8Q |
64 MByte |
Est 3 MByte per core |
|||
256 port QSFP28 100 Gb/s |
8Q |
64 MByte |
Est 3 MByte per core |
|||
32 port QSFP28 100 Gb/s |
8Q |
16 MByte |
Est 3 MByte per core |
|||
32 port QSFP28 OR 64 QSFP |
8Q |
16 MByte |
Est 3 MByte per core |
|||
48 SFP28 25 Gb/s and 6 QSFP28 100 G |
8Q |
16 MByte |
Est 3 MByte per core |
|||
32 QSFP28 100 Gb/s |
8Q |
16 MByte |
Est 3 MByte per core |
|||
32 QSFP28 100 Gb/s |
8Q |
16 MByte |
Est 3 MByte per core |
|||
32 QSFP28 100 Gb/s |
8Q |
16 MB |
Est 3 Mbyte per core |
|||
32 QSFP28 100 Gb/s |
||||||
32 QSFP28 100 Gb/s |
8Q |
16 MB |
Est 3 Mbyte per core |
|||
48 x SFP28 + 8 x QSFP28/s |
8Q |
16 MB |
Est 3 Mbyte per core |
|||
6 QSFP28 100 Gb/s and 48 SFP28 25 Gb/s |
8Q |
16 MB |
Est 3 Mbyte per core |
|||
32 QSFP28 100 Gb/s |
8Q |
16 MB |
Est 3 Mbyte per core |
|||
128 QSFP28 100 Gb/s |
8Q |
256 MB |
Est 3 Mbyte per core |
|||
16 QSFP28 AND 8 200 Gb/s coherent |
8Q |
Est 3 Mbyte per core |
||||
48 SFP28 25 Gb/s and 6 QSFP28 100 Gb/s |
8Q |
16 MB |
||||
48 SFP28 25 Gb/s and 6 QSFP28 100 Gb/s |
8Q |
16 MB |
||||
48 SFP28 and 6 QSFP28 |
8Q |
?? MB |
Est 4.5 Mbyte per core |
|||
48 SFP28 25 Gb/s and 6 QSFP28 100 Gb/s |
8Q |
?? MB |
||||
32 QSFP28 100 Gb/s AND 2 SFP+ |
8Q |
22 MB |
Est 4.5 Mbyte per core |
|||
48 SFP28 25 Gb/s AND 6 QSFP28 |
8Q |
22 MB |
Est 4.5 Mbyte per core |
|||
64 QSFP 40 Gb/s AND 2 SFP+ |
8Q |
16 MB |
Est 3 Mbyte per core |
|||
64 QSFP28 100Gb/s AND 2 SFP+ |
8Q |
42 MB |
10.5 Mbyte per slice |
|||
64 QSFP28 100 Gb/s AND 2 SFP+ |
8Q |
16 MB |
Est 3 Mbyte per core |
|||
64 QSFP28 100 Gb/s AND 2 SFP+ |
8Q |
42 MB |
Est 10.5 Mbyte per slice |
|||
64 QSFP28 100 Gb/s AND 2 SFP+ |
8Q |
42 MB |
Est 10.5 Mbyte per slice |
|||
32 x QSFP-DD 400 Gb/s ports |
||||||
32 x OSFP 400 Gb/s ports |
64 MB |
|||||
32 x QSFP-DD 400 Gb/s ports |
64 MB |
|||||
Innovium Teralynx Hot new chip to take over the world [2017 OCP] | ||||||
32 x QSFP-DD 400 Gb/s ports + 2 x SFP |
70 MB |
|||||
8-slot chassis w 4-port 400G and 16-port 100G cards |
70 MB |
|||||
Broadcom Jericho a member of the StrataDNX DUNE family -- includes Qumran | ||||||
288 QSFP28 100 Gb/s ports |
4 GByte per ASIC |
VoQ 10 mS per queue |
||||
six QSFP28 100 Gb/s ports and 48 x 10 Gb/s |
4 GByte per ASIC |
VoQ 10 mS per queue |
||||
36 x 100 Gb/s QSFP28 linecard |
4 GByte per ASIC |
VoQ |
||||
36 x 40 Gb/s QSFP linecard |
4 GByte per ASIC |
VoQ |
||||
36 x QSFP28 100G |
8Q |
24 GB |
||||
30 x 40G QSFP and 6 x 100G |
8Q |
8 GB |
||||
48 port sfp+ and 2 x 100G |
8Q |
4 GB |
||||
48 x 10 G SFP+ and 6 x QSFP28 |
8Q |
4 GB | 4 GB VoQ |
|||
48 x 10GBASE-T and 6 x QSFP28 |
8Q |
4 GB | 4 GB VoQ |
|||
24 QSFP+ and 12 QSFP28 |
8Q |
8 GB | 8 GB in 2 groups |
|||
48 QSFP+ and 24 QSFP28 |
8Q |
32 GB | 32 GB in 8 groups |
|||
36 QSFP28 per linecard |
8Q |
4 or 6 GB per 6-port group VoQ |
||||
6 QSFP28 100/40 Gb/s and 48 10/1 Gb/s |
8Q |
6 GB | 6 GB VoQ |
|||
48 SFP+ and 6 QSFP28 |
8Q |
4 GB | ||||
48 SFP+ and 6 QSFP28 |
8Q |
4 GB | ||||
48 SFP+ and 6 QSFP28 |
8Q |
4 GB | ||||
48 SFP+ and 6 QSFP28 |
8Q |
4 GB | ||||
modules: 6 x 100G, 16 x 40G, 24 x 10G |
|
|||||
48 SFP+ and 6 QSFP28 |
8Q |
6 GB | ||||
Broadcom Jericho+ a member of the StrataDNX DUNE family | ||||||
Agema AGC5648 Jericho+ |
48 SFP28+ and 6 QSFP28 |
8Q |
8 GB per ASIC | |||
36 ea QSFP28 100 Gb/s |
8Q |
4 GB per ASIC |
||||
48 x 25 G SFP28 and 6 x QSFP28 |
8Q |
8 GB | 8 GB VoQ |
|||
36 x QSFP28 100 Gb/s |
8Q |
4 GByte per ASIC | ||||
Broadcom Jericho2 a member of the StrataDNX DUNE family | ||||||
Broadcom ARAD (July 2014) a member of the StrataDNX DUNE family | ||||||
48 port SFP+ and 4 x QSFP |
8Q |
9 GB |
VoQ 125 MB per 10 gig port |
|||
48 port SFP+ and 2 x QSFP28 100 Gb/s |
8Q |
9 GB |
VoQ 125 MB per 10 gig port |
|||
48 port SFP+ and 2 x MXP |
8Q |
9 GB |
VoQ 125 MB per 10 gig port |
|||
48 port sfp+ line card |
8Q |
3 GB per each of 3 processors |
50 MB/port-queue |
|||
12 QSFP28 100G |
8Q |
3 GB per each of 6 processors |
500 MB/port-queue |
|||
24 SFP+ and 4 QSFP+ |
?? |
12 GByte |
VoQ very large |
|||
32 QSFP28 100 Gb/s |
?? |
24 MB |
||||
6 QSFP28 100 Gb/s and 48 SFP28 |
?? |
24 MB |
||||
32 QSFP28 100 Gb/s |
?? |
24 MB |
||||
32 QSFP28 100 Gb/s |
?? |
24 MB |
||||
32 QSFP28 100 Gb/s |
?? |
24 MB |
||||
6 QSFP28 100 Gb/s and 48 QSFP28 |
?? |
24 MB |
||||
6 QSFP28 100 Gb/s and 48 QSFP28 |
?? |
24 MB |
||||
32 x QSFP28 100G |
16 MByte | 16 Mbyte | ||||
65 x QSFP28 100G |
16 MByte | 16 Mbyte | ||||
48 x SFP28 25G and 6 x QSFP28 100G |
16 MByte | 16 Mbyte | ||||
32 x QSFP28 and 2x SFP+ |
20 MByte | 20 Mbyte | ||||
64 x QSFP28 |
22 MByte | 20 Mbyte | ||||
48 x SFP28 and 6 x QSFP28 |
20 MByte | 20 Mbyte | ||||
32 x QSFP28 |
22 MByte | 20 Mbyte | ||||
2 x QSFP28 100G and 48 x SFP+ 10G |
||||||
|
||||||
48 SFP+ and 2 QSFP and 4 QSFP expansion |
?? |
8 MB to 1 port |
||||
24 SFP+ and 2 QSFP and 4 QSFP expansion |
?? |
8 MB to 1 port |
||||
|
||||||
48 SFP+ and 12 QSFP+ |
?? |
4.6 MB |
64 KB to 1 port |
|||
36 QSFP+ |
?? |
4.6 MB |
128 KB to 1 port |
|||
32 QSFP28 100Gb/s |
?? |
16 MB |
12 MB to 1 port? |
|||
|
||||||
48 1000-base-T + 4 SFP+ |
8Q |
4 MB/switch |
4 MB |
|||
(24+2) 1000-base-T and 4 SFP+ |
8Q |
2 MB |
2 MB to 1 port | |||
(24+2) 1000-base-T and 4 SFP+ |
8Q |
2 MB |
2 MB to 1 port | |||
48 x 1000bsseT and (2 x 40G or 4 x 10G front module) |
8Q |
5 MB |
||||
24 x 1000BaseT and 24 multiGig ports and (1 x 100G or 2 x 40G front module) |
8Q |
12 MB |
||||
48 1000-base-T and 4 SFP+ and 2 QSFP stacking ports |
8Q |
4 MB |
? MB to 1 port | |||
48 1000-base-T and 4 SFP+ and 2 QSFP stacking ports |
8Q |
4 MB |
||||
48 1000-base-T and 4 SFP+ |
8Q |
4 MB |
? MB to 1 port | |||
48 1000-base-T w 2 SFP+ |
8Q |
4 MB |
? MB to 1 port | |||
|
||||||
20 x QSFP 40G and 4 x SFP+ and 4 x QSFP28 100G |
9 MB |
|||||
48 x SFP+ 10G and 2 x QSFP+ 40G and 4 x QSFP28 100G |
9 MB |
|||||
48 x 1000baeT and 4 x QSFP+ 40G and 4 x QSFP28 100G |
3 MB |
|||||
48 x SFP 1G and 4 x SFP+ 10G |
6 MB |
|||||
|
||||||
Catalyst 3850, 3650 |
ASIC support 24 GE ports AND 2 x 10GE |
8Q |
several |
6 MB per ASIC |
4 MB per ASIC |
|
Catalyst C3850-12XS |
12 SFP+ (6 ports per ASIC) |
8Q |
2P6Q3T |
6 MB per ASIC |
4 MB per ASIC |
|
ASIC support 24 GE ports AND uplink module |
8Q |
2P6Q3T | ||||
48 GE ports AND uplink module |
8Q |
2P6Q3T | ||||
24 100/1G/2.5G/5G/10G Tw-Pr ports AND uplink module |
8Q |
2P6Q3T | ||||
24 QSFP 40 Gb/s |
8Q |
2P6Q3T | ||||
24 SFP+ ports AND uplink module |
8Q |
2P6Q3T | ||||
24 SFP28 ports AND 4 QSFP28 100G |
8Q |
2P6Q3T | 36 MB per ASIC |
est 23 MB |
||
16 x QSFP28 100G OR 32 QSFP 40G |
8Q |
2P6Q3T | 36 MB per ASIC |
est 23 MB |
||
32 SFP28 100G ports |
8Q |
2P6Q3T | 36 MB per ASIC |
est 23 MB |
||
48 SFP+ ports AND 4 QSFP28 100G |
8Q |
2P6Q3T | 36 MB per ASIC |
est 23 MB |
||
Other Shared Memory |
||||||
24 10/100-base-T + 2 x 1G GBIC |
8Q |
2 MB |
2 MByte |
|||
2 x SFP+ and 4 SFP and 4 SFP-or-TwPr (10/100/1000) |
8Q |
|
||||
48 1000-base-T + 4 x 10G SFP+ |
8Q |
4 MB |
? |
|||
16 10-Gbps SFP+ |
8Q |
2 MB |
2 MB |
|||
10 10-Gbps SFP+ |
8Q |
2 MB |
2 MB |
|||
48 SFP28 and 6 QSFP28 |
Single ASE3 20 MByte |
10.2 MB |
||||
2RU 72 x 40Gb/s QSFP+ |
Single ASE2 30 Mbyte |
5.1 MB |
||||
48 10-Gbps BASE-T and 4 QSFP+ |
12 MB NFE + 25 MB ALE |
21 MB max burst |
||||
48 SFP+ and {12 QSFP+ or 4 CPAK 100G} |
12 MB NFE + 40 MB ALE |
21 MB max burst |
||||
48 SFP28 25 Gb/s and 6 QSFP28 |
40 MByte |
17.6 MB |
||||
48 tw-pr 10 Gb/s and 6 QSFP28 |
40 MByte |
17.6 MB |
||||
16 QSFP-DD 400/100 Gb/s |
80 MByte |
|||||
28 SFP28 and 8 QSFP28-DD |
80 MByte |
|||||
Brocade FCX624S |
24Gig-E and 4 SFP+ |
8Q |
2 MB |
1.04 MB to 1 port |
||
40 SFP+ plus 8 SFP+ |
8Q |
4 MB per ASIC |
230 KBytes |
|||
24Gig-E w 8 SFP+ |
8Q |
4 MB |
1 MB to 1 port | |||
24Gig-E and 4 10Gb/s |
8Q or 4Q or 2 Q |
0.23 MB |
||||
24 tw-pr Gig-E and 4 optional 10Gb/s |
8Q |
11.25 MB |
4.5 MB shared |
6.75 MB shared |
||
20 tw-pr Gig-E and 4 SFP |
4Q |
0.512 MB shared |
0.512 MB shared |
|||
HP A5800-24G |
24Gig-E and 4 10Gb/s |
8Q |
4 MB |
4 MB |
||
HP E6600-24G-4XG |
24Gig-E w 2 10Gb/s |
8Q |
18 MB for GE |
18 MB |
||
24 x SFP+ |
8Q |
2 MB/sw-chip |
shared |
1.238 + 0.02 MB |
||
48 SFP+ |
8Q |
2 MB/sw-chip, 8 MB total |
1.238 + 0.02 MB |
|||
24 SFP+ |
8Q |
|||||
52 SFP+ |
8Q |
|||||
48 SFP+ and 4 QSFP+ |
8Q |
|||||
24 SFP+ |
2Q |
2 MB |
||||
48 Gig-E, 4 x SFP+ |
8Q |
0.75 MByte |
98 KB per port |
|||
24 Gig-E, 4 x SFP+ |
8Q |
4 MB |
4 MB |
|||
48 GE + 4 SFP+ |
8Q |
4 MB/switch |
4 MB |
|||
48 GE + 4 SFP shared |
8Q fixed |
786 KB per 12-port |
131 KB w 8 Queues |
|||
48 GE + 4 SFP shared |
1 to 8Q |
786 KB per 12-port |
786 KB w 1 Queue |
|||
Catalyst 3750G-48TS, 2960G |
12 ASICs w/ 4 GE ports ea |
8Q |
1P3Q3T |
576KB per ASIC |
192KB per ASIC |
384 KB per ASIC |
Catalyst 2960S |
Single ASIC: 48 +2 10Gb/s |
8Q |
1P3Q3T |
2 MByte |
2 MByte |
|
Catalyst 3750E, 3560E, 3750X & 3560X |
ASIC support 24 GE ports or 2 x 10G |
2Q |
1P3Q3T |
2750KB per ASIC |
750KB per ASIC |
2 MB per ASIC |
Catalyst 2960-X |
4Q |
4 MB per ASIC |
Catalyst 4948E |
48 GigE + 4 SFP+ |
4 |
17.5 MB |
Max 16 MB to 1 port |
Cisco Ctalyst 4510R+E |
Chassis w 8 interface slots -- 12 port SFP+ line card |
8 |
1P7Q1T |
32 MB on Supervisor |
||
Cisco Catalyst WS-C4500X-32SFP+ |
1 RU w 32 SFP+ ports and module for 8 more |
8 |
1P7Q1T |
32 MB |
see note above |
|
48 SFP+ |
4 |
18 MB |
5.8 MB |
|||
48 Gig-E + 4 SFP+ |
4Q |
1250 MB |
enough |
|||
48 Gig-E + 4 SFP+ |
8Q |
768 MB |
16 MB/port? |
GOBBS |
||
32 1/10 G SFP+ |
8Q |
6 GB |
||||
32 1/10 G SFP+ and 2 QSFP28 100G |
8Q |
6 GB |
||||
32 1/10 G SFP+ and 2 QSFP28 100G + stack |
8Q |
6 GB |
||||
24 SFP+ and 2 CFP/2 100G |
8Q |
20 GB |
||||
48 SFP+ and 4 CFP/2 100G |
8Q |
20 GB |
||||
Virtual Output Port queuing |
||||||
8Q |
3 GB | 256 MB/port-queue |
||||
8-port 10 Gb/s line card |
8Q |
3 GB | 256 MB/port-queue |
|||
24-port 1 Gb/s line card |
8Q |
1 GB | 64 MB/port-queue |
|||
96-port SFP+ line card 48-port QSFP+ line card 4-port CFP2 100 Gb/s line card |
8Q |
9 MB per 24 10-Gb/s ports 9 MB per 6 40-Gb/s ports 36 MB per fabric module | ||||
26 SFP+ |
8 |
1 (I think) |
480 KB per port |
480 KB |
||
32 fixed SFP+ and 16-port expansion module |
8 Unicast 8 Multi |
1 (I think) |
640 KB per port |
640 KB |
||
48 fixed SFP+ and 6 QSFP+ |
8 Unicast 8 Multi |
1 (I think) |
25 MB per 12-port 10G ASIC |
14 MB shared per ASIC |
3 MB Unicast + 6 MB Multicast |
|
48 fixed SFP+ and 4 QSFP+ |
8 Unicast 8 Multi |
1 (I think) |
25 MB per 12-port 10G ASIC |
14 MB shared per ASIC |
3 MB Unicast + 6 MB Multicast |
|
Chassis solution. 1-6 ports per ASIC |
ASICs have 0.5 to 3.3 GB/port |
|||||
QSFP: 12 x 100 Gb/s OR 36 x 40 Gb/s |
8Q |
4 GByte per ASIC |
VoQ 100 mS per port |
|||
Cisco Nexus 7000 (M1 series) |
48-port Tw-Pr GE |
2q4t |
1p3q4t |
7.56 MB per port |
6.15 MB per port | |
Cisco Nexus 7000 (F3 Series) |
6-port 100GE CPAK |
4q |
4q |
72 MB shared |
||
Cisco Nexus 7000 (F2 series) |
48 SFP+ across 12 ASICs |
4q1t |
1p3q1t |
72 MB shared |
||
Cisco Nexus 7000 (F3 series) |
12 QSFP+ across 6 ASICs |
4q1t |
1p3q1t |
72 MB shared |
||
Cisco Nexus 7000 (M1 series) |
32-port 10GE shared 4-port groups SFP+ |
8q2t |
1p7q4t |
65 MB per group |
80 MB per group | |
Cisco Nexus 7000 (M1 series) |
8-port 10GE X2 |
8q2t |
1p7q4t |
92 MB |
80 MB | |
Cisco Nexus 7000 (M2 series) |
2-port 100GE CFP2 |
8q2t |
1p7q4t |
62.46 MB |
31.23 MB | |
Cisco Nexus 7000 (M2 series) |
24-port 10GE SFP+ |
8q2t |
1p7q4t |
5.21 MB |
5.21 MB | |
Cisco Nexus 7700 (F3 Series) |
12-port 100GE CPAK across 12 ASICs |
4q |
4q |
144 MB shared |
||
Cisco Nexus 7700 (F3 Series) |
24-port 40GE QSFP+ across 12 ASICs |
4q |
4q |
144 MB shared |
||
Cisco Nexus 7700 (F3 Series) |
48-port 10GE SFP+ across 6 ASICs |
4q |
4q |
144 MB shared |
||
Cisco Catalyst 6500 gets its own table |
Comments, suggestions, corrections to warner@ucsc.edu