Optimize Google and Netflix with ISP

This is the first part of a series to explore, regarding optimizing enterprise Internet applications.

I estimated that on average, I use Google at least a dozen times on an average work day. I don’t think I will be far off to say that it is the top Internet application in almost every enterprise. While writing this article, I searched a couple of key words and arrived at a few web sites to find the specific information I was looking for. I did not have any books or references with me. More often, business users expect information on demand. It is often faster to get the information on the Internet than finding on the hard drive, and sometimes more accurate than asking your buddies.

Why this matters? Google is just an example here. For an Internet application that triggers millions of transactions almost every day in your enterprise, it is worthwhile to look “into the cloud”, and understand how your enterprise is interconnected to the rest of the world.

Google’s primary AS is 15169. It originates more than a hundred IPv4 prefixes for common apps, such as www.google.com. So how are you connected to AS15169? The average AS path length from Google is almost 3. But if you share one of the ISPs that Google peer directly with, then your path will be shorter.

The following lists the top IPv4 peers for AS 15169:

ASN Name

AS3356 Level 3 Communications

AS3549 Global Crossing

AS1299 TeliaNet Global Network

AS7018 AT&T Services, Inc.

AS174 Cogent Communications

AS6762 Telecom Italia Sparkle

AS3320 Deutsche Telekom AG

AS286 KPN Internet Backbone

AS6939 Hurricane Electric, Inc.

AS3257 Tinet SpA

As an example, check the internet routing table to see how you reach IP address for Google.com. The second shorter path (through a tier one ISP) will always be preferred over the first one.

Internet_router#show ip bgp 173.194.33.104
BGP routing table entry for 173.194.33.0/24, version 2010015
Paths: (2 available, best #2, table default)
Not advertised to any peer
13789 2828 15169
…
3356 15169
…

The significance goes beyond the length of the path or how many ISPs to transit through. Larger ISPs are becoming content providers in order to further optimize performance. Just a few days ago, Level 3 signed a deal with Netflix to “support storage for the entire Netflix library of content”. Level 3 has become the primary content network for Netflix, while Akamai and Limelight remain as partners. In this case, if your users are in the entertainment business, you will be at significant advantage by peering directly with Level 3.

There are more considerations, inbound vs outbound, streaming, CDN, geographical, international, resilience and cost… more to follow.

BGP timer and Cisco/Juniper negotiation

Juniper SRX integrates firewall features with full routing capabilities. In a network architecture where there are increasing demand for security segmentation and stateful firewall inspection within the network, multi-vender routing interaction may become necessary.

Risks and complexity exist with a mixed vendor environment, even when both sides support standard protocols. Here is such an example.

To establish BGP neighbor relationship, both sides need to agree on timer values. RFC4271 defines the characteristics of the timer, but does not specify value for keepalive and hold time.

BGP uses keepalive timers to monitor established connections. If for a period of time that exceeds hold time, BGP considers the neighbor connection down. Note the hold time is always three times the keepalive time.

By default, vendors set timers differently:
vendor Keepalive (second) Hold time (second)
Cisco Nexus (4.2) 60 180
Juniper SRX (10.0) 30 90

The timers should match between vendors. This is based on configuration, or negotiated behavior. Note with Juniper SRX, BGP on the local routing device uses the smaller of either the local hold-time value or the peer’s hold-time value received in the open message as the hold time for the BGP connection between the two peers. Therefore, by setting timer on Cisco Nexus to a smaller value (10/30), it is used by SRX.

To reduce convergence, the test sets Nexus keepalive to 10 seconds. This shows after successful negotiation, SRX (with a local hold time of 90) uses Nexus’s keepalive of 10 seconds, and active hold time of 30 seconds.

On Juniper SRX:
Peer: 10.88.15.14+179 AS 65000 Local: 10.88.14.4+54104 AS 65000
Type: Internal State: Established Flags:
Last State: OpenConfirm Last Event: RecvKeepAlive
Last Error: None
Export: [ bgp_outbound_policy ]
Options:
Local Address: 10.88.14.4 Holdtime: 90 Preference: 170
Number of flaps: 0
Peer ID: 10.88.120.130 Local ID: 10.88.14.4 Active Holdtime: 30
Keepalive Interval: 10 Peer index: 0
…

On Cisco Nexus:
Nexus7k# sh ip bgp nei vrf test
BGP neighbor is 10.88.14.4, remote AS 65000, ibgp link, Peer index 1
BGP version 4, remote router ID 10.88.14.4
BGP state = Established, up for 00:53:15
Last read 00:00:05, hold time = 30, keepalive interval is 10 seconds
Last written 0.956168, keepalive timer expiry due 00:00:09
Received 20433 messages, 1 notifications, 0 bytes in queue
Sent 18392 messages, 0 notifications, 0 bytes in queue
Connections established 2, dropped 1
Last reset by peer 02:05:25, due to peer deconfigured
Last reset by us never, due to process restart

Neighbor capabilities:
Dynamic capability: advertised (mp, refresh, gr)
Dynamic capability (old): advertised
Route refresh capability (new): advertised received
Route refresh capability (old): advertised received
4-Byte AS capability: advertised received
Address family IPv4 Unicast: advertised received
Graceful Restart capability: advertised received
…

Enable BGP load sharing with Multipath-relax

If your network is like most large enterprise’s, chances are you have Internet connection from dual ISPs to your data center.

It used to be a popular tactic to have a primary provider, and a backup provider which you pay lesser bandwidth for. In recent years, usage based circuits has given way to flat rate, often larger bandwidth circuits from both ISPs. After all, users demand more bandwidth, even in failure scenarios.

Why would you pay for expensive Internet access and only letting it sitting idle most of the time? Therefore, load sharing on redundant ISP connections has become highly desirable for data centers.

By default, BGP only installs one route into the routing table. To accomplish load sharing, BGP multipath needs to be enabled. To further complicate the matter, BGP requires all paths to have the following equal characteristics in order to become equal cost path:
• Weight
• Local preference
• AS-PATH length
• Origin
• MED
• AS-PATH

Since neighboring AS or AS-PATH must be equal, that means BGP can only load balance to a single AS or single ISP. Since different ISP will have different AS and AS-PATH, BGP will only pick one best route.

So how does Cisco support load balancing to two ISPs? There is a hidden command bgp bestpath as-path multipath-relax

I don’t know why it is hidden command, since it appears to be a very useful feature, and it is supported. But do use with caution, as the number of routing entries will increase, often overwhelming the limited memory you have on the Internet router.

In this year’s Networker, an alternative method was also shown, essentially separating out the Internet routing table by even and odd digits, for outbound traffic.

Not all ports are created equal on Nexus5000

With Nexus 5000, it is very common to select the first few available ports for basic infrastructure such as peer links and uplinks. But that may not be best practice for resiliency, due to shared use of ASIC on the switch.

In short, there are 5 ASICs. Port e1/1-4 share the first ASIC, for example. If the first 4 physical port on Nexus 5000 are selected for all peer link and uplink to Nexus 7000, then loss of ASIC0 will bring down all these links at the same time. Even with the protection of VPC which can minimize loss, still not a desirable scenario.

Use “show hardware internal gatos all-ports” to see port to ASIC mapping. Below shows "gat" 0 which is the ASIC number for e1/1-4. It will also show the next four ports show the next ASIC, so on so forth.

Nexus-5010-sw1# show hardware internal gatos all-ports
process_cli_req:
Calling Hander:
Gatos Port Info:
name |log|gat|mac|flag|adm|opr|c:m:s:l|ipt|fab|xgat|xpt|if_index|diag
-------+---+---+---+----+---+---+-------+---+---+----+---+--------+----
xgb1/4 |3 |0 |0 |b7 |en |up |0:0:0:f|0 |55 |0 |2 |1a003000|pass
xgb1/3 |2 |0 |1 |b7 |en |up |0:1:1:f|1 |54 |0 |0 |1a002000|pass
xgb1/1 |0 |0 |2 |b7 |en |up |1:2:2:f|2 |56 |0 |4 |1a000000|pass
xgb1/2 |1 |0 |3 |b7 |en |up |1:3:3:f|3 |57 |0 |6 |1a001000|pass
xgb1/8 |7 |1 |0 |b7 |en |up |0:0:0:f|0 |50 |1 |2 |1a007000|pass
xgb1/7 |6 |1 |1 |b7 |en |up |0:1:1:f|1 |51 |1 |0 |1a006000|pass
xgb1/5 |4 |1 |2 |b7 |en |up |1:2:2:f|2 |53 |1 |4 |1a004000|pass
xgb1/6 |5 |1 |3 |b7 |en |up |1:3:3:f|3 |52 |1 |6 |1a005000|pass
xgb1/19|18 |2 |0 |b7 |en |up |0:0:0:f|0 |45 |4 |0 |1a012000|pass
xgb1/11|10 |2 |1 |b7 |dis|dn |0:1:1:f|1 |44 |2 |0 |1a00a000|pass
xgb1/20|19 |2 |2 |b7 |en |up |1:2:2:f|2 |49 |4 |2 |1a013000|pass
xgb1/12|11 |2 |3 |b7 |dis|dn |1:3:3:f|3 |48 |2 |2 |1a00b000|pass
xgb1/18|17 |3 |0 |b7 |en |up |0:0:0:f|0 |41 |4 |6 |1a011000|pass
xgb1/10|9 |3 |1 |b7 |en |up |0:1:1:f|1 |40 |2 |6 |1a009000|pass
xgb1/17|16 |3 |2 |b7 |en |up |1:2:2:f|2 |46 |4 |4 |1a010000|pass
xgb1/9 |8 |3 |3 |b7 |en |up |1:3:3:f|3 |47 |2 |4 |1a008000|pass
xgb1/15|14 |4 |1 |b7 |en |up |0:1:1:f|1 |36 |3 |0 |1a00e000|pass
xgb1/16|15 |4 |3 |b7 |en |up |1:3:3:f|3 |42 |3 |2 |1a00f000|pass
sup0 |20 |5 |0 |b7 |en |dn |0:0:0:0|0 |33 |0 |0 |15020000|pass
sup1 |21 |5 |1 |b7 |en |dn |0:0:0:1|0 |33 |0 |0 |15010000|pass
xgb1/14|13 |5 |1 |b7 |en |up |0:1:1:f|1 |32 |3 |6 |1a00d000|pass
xgb1/13|12 |5 |3 |b7 |en |up |1:3:3:f|3 |38 |3 |4 |1a00c000|pass
mfc2/5 |4 |6 |0 |b7 |dis|dn |0:0:0:0|0 |31 |6 |0 |01084000|pass
mfc2/6 |5 |6 |1 |b7 |dis|dn |0:0:0:1|0 |31 |6 |1 |01085000|pass
mfc2/7 |6 |6 |2 |b7 |dis|dn |0:1:0:2|1 |30 |6 |2 |01086000|pass
mfc2/8 |7 |6 |3 |b7 |dis|dn |0:1:0:3|1 |30 |6 |3 |01087000|pass
mfc2/4 |3 |6 |4 |b7 |dis|dn |1:2:2:0|2 |35 |5 |6 |01083000|pass
mfc2/3 |2 |6 |5 |b7 |dis|dn |1:2:2:1|2 |35 |5 |4 |01082000|pass
mfc2/2 |1 |6 |6 |b7 |dis|dn |1:3:2:2|3 |34 |5 |2 |01081000|pass
mfc2/1 |0 |6 |7 |b7 |dis|dn |1:3:2:3|3 |34 |5 |0 |01080000|pass

If you use dual physical ports for peer link, then definitely use two ports that are different ASICs.

Why “gatos”? Here is a little humor uncovered by Colin McNamara you might enjoy.

Nexus7000 OSPF failure due to MFDM crash – still searching for root cause

This occurred a while ago, still waiting for words. Unfortunately the condition has cleared due to production network, just wondering if anybody else out there has some possible clue about root cause.

The most noticeable symptom was OSPF adjacency problem, with no error message. OSPF was able to establish at least partial adjacency with some neighbors, but not others.

Thinking OSPF process has gone bad, we restarted it, but no use. OSPF process runs fine with no error, but adjacency trouble remains. Further review of logs reveals that a process known as MFDM (Multicast FIB Distribution?) crashed with attempted to restart three times but did not recover. Obviously OSPF adjacency uses Multicast FIB (MFIB).

2010 Aug 2 17:29:57 Nexus-7010 %SYSMGR-2-SERVICE_CRASHED: Service "mfdm" (PID 16377) hasn't caught signal 11 (core will be saved).
2010 Aug 2 17:30:03 Nexus-7010 %SYSMGR-2-SERVICE_CRASHED: Service "mfdm" (PID 16471) hasn't caught signal 11 (core will be saved).
2010 Aug 2 17:30:03 Nexus-7010 %SYSMGR-2-SERVICE_CRASHED: Service "mfdm" (PID 16524) hasn't caught signal 11 (core will be saved).

We could not recover MFDM individually, end up reloading VDC to recover it. Magically, OSPF started working again!

Still an open ticket, no root cause has been identified. Part of the difficulty was we seem to have lost some of the files, which makes it hard for vendor to trace it down. Just wondering is there is any similar experience out there?

And, could a hardware or ASIC related failure be causing MFDM crash? Or is it more likely a software bug?

Thanks for your thoughts and comments.

Nexus7000/5000/1000v – A closer look at VPC and port channel load balancing

Port channel is a great redundacy and load sharing feature in data centers. Cisco Nexus takes it one step further with Virtual Port Channel (VPC). There are numerous good documentations about VPC, including the downloadable design guide.
Sometimes you need to look under the cover, and see exactly how each physical port is utilized by port channels. This note highlights a few useful commands.

Strictly speaking, port channel does load sharing, not load balancing. So there should not be an expectation for 50/50 balance. How well load sharing works largely depends on the traffic and the hashing method selected. For example, Nexus 1000v supports 17 hashing algorithms to load-share traffic across physical interfaces in a PortChannel, including source-based hashing and flow-based hashing. The default is source-mac.

Another important concept: hashing is uni-directional, determined by the sending party. Therefore there is no guarantee that load sharing will be symmetrical. For illustration, Nexus 7k and 5k are connected in what is known as “back to back” VPCs (port channel 75). A Nexus 7k has multiple physical connections southbound on the same logical channel. It determines the physical port to send traffic based on local hashing, as illustrated by the green arrow.

Here is the first command which shows the hashing algorithem used:
show port-channel load-balance
N-7010-1# sh port-c load
Port Channel Load-Balancing Configuration:
System: source-dest-ip-vlan
Port Channel Load-Balancing Addresses Used Per-Protocol:
Non-IP: source-dest-mac
IP: source-dest-ip-vlan

The second command shows how well load sharing is working on your port channels. Note statistics are accumulative, and is reset by clearing corresponding interface counters.
Show port-channel traffic

N-7010-1# show port-c traffic
ChanId Port Rx-Ucst Tx-Ucst Rx-Mcst Tx-Mcst Rx-Bcst Tx-Bcst
------ --------- ------- ------- ------- ------- ------- -------
75 Eth1/5 53.16% 42.58% 49.91% 53.70% 44.03% 44.86%
75 Eth1/6 46.83% 57.41% 50.08% 46.29% 55.96% 55.13%

As in the diagram, orange arrow indicates Nexus 5k northbound load sharing based on its hashing algorithem. The blue arrow indicates Nexus 1kv northbound load sharing based on its hashing.

Since Netflow is not yet supported on Nexus 5k, how do we tell which physical interface will a certain flow take? Here is the third command:

show port-channel load-balance forwarding-path interface ...

The following example shows how different IP address pairs yields different port utilization.
N-5010-1# show port-channel load-balance forwarding-path interface port-channel 75 vlan 25 src-ip 10.17.19.15 dst-ip 10.17.21.15
Missing params will be substituted by 0's.
Load-balance Algorithm on switch: source-dest-ip
crc8_hash: 90 Outgoing port id: Ethernet1/3
Param(s) used to calculate load-balance:
dst-ip: 10.174.21.15
src-ip: 10.174.19.15
dst-mac: 0000.0000.0000
src-mac: 0000.0000.0000

N-5010-sw1# show port-channel load-balance forwarding-path interface port-channel 75 vlan 25 src-ip 10.17.19.10 dst-ip 10.17.34.15
Missing params will be substituted by 0's.
Load-balance Algorithm on switch: source-dest-ip
crc8_hash: 179 Outgoing port id: Ethernet1/4
Param(s) used to calculate load-balance:
dst-ip: 10.174.34.15
src-ip: 10.174.19.10
dst-mac: 0000.0000.0000
src-mac: 0000.0000.0000

If you are running a test, and notices that the traffic is not well balanced, now you have a method to check the physical port allocation for your particular end points. You also have the option to experimenting with different hashing algorithem to suit your needs.

Nexus1000v – when to use “system mtu”

I can’t be the first one confused about jumbo, mtu, and system mtu on Nexus 1000v. After reading some excellent posts, all signs were indicating that “system mtu” was designed to solve the “chicken and egg” problem of running VSM on IP storage.

Like "system vlan", “system mtu” applies to system uplink profile only. So if VSM is not even running under VEM (it runs on vSwitch), there is no need to set “system mtu”, right?

Well, not quite. It turns out system mtu is still needed to preserve the connection to VEM. Assuming jumbo is used (for storage as an example), reboot of ESX will revert the physical NIC to default MTU (1500), which results in mismatched MTU between physical NIC and virtual NIC, and loss of connectivity. “system mtu” preserves the setting on physical NIC, and thus preventing VEM from disappearing.

To further clarify, here is an example of configuring jumbo of 9000 on Nexus 1000v
1. “system jumbomtu 9000” (global)
2. “system mtu 9000” (uplink port profile)

That is all. Note once set, “system mtu” overwrites “mtu”, therefore there is no need to set interface mtu explicitly.

A couple of things potentially confusing:
-The show commands on Nexus 1000v is not exactly accurate for MTU, fix is supposed to be coming
-There is an error in Cisco Nexus command reference, which states “The value that is configured for system mtu command must be less then value configured in the system jumbomtu command”. It should be “less or equal to”; There are no reason to set system mtu to 8998 unless hardware dictates so.

I hope that clear up some confusions. If you notice any behavior inconsistent with this understanding, please kindly let me know.

Nexus 1000v bug - Three ways to check VSM HA health

If you are using Nexus 1000v, more than likely you have set up VSM high availability. To ensure system stability and gain the true benefit of HA, check if VSM HA is truly synchronized. Don’t stop when you see 2 supervisor modules. Otherwise you may be caught at the worst time of a failure scenario, and found out there is no HA, and you have to deal with a bug, even configuration loss.

Specifically, check after initial setup, and after system operations such as VSM reload. There is a bug, CSCtg46327, fixed in 4.0(4)SV1(3a). It prevents VSM active and standby to synchronize. Standby continues to come up and down in attempts to do so.

You can check for this bug and other potential VSM HA problem using these methods:

1. “show module” output affected by the bug, module 2 should show “VSM, Nexus 1000V, ha-standby”, instead of “powered-up”:

vsm# sh module
Mod Ports Module-Type Model Status
--- ----- -------------------------------- ------------------ ------------
1 0 Virtual Supervisor Module Nexus1000V active *
2 0 Supervisor/Fabric-1 powered-up
5 248 Virtual Ethernet Module NA ok
6 248 Virtual Ethernet Module NA ok
7 248 Virtual Ethernet Module NA ok
...

2. “show svs neighbor” list standby VSM MAC as a “VEM”, which is incorrect. It should be type VSM:
vsm# sh svs nei
Active Domain ID: 91
AIPC Interface MAC: 0050-56b4-52bb
Inband Interface MAC: 0050-56b4-3fc1


Src MAC Type Domain-id Node-id Last learnt (Sec. ago)
------------------------------------------------------------------------
0050-56b4-5eb5 VEM 0 ffffffff 0.00
0002-3d43-8504 VEM 901 0502 160591.40
0002-3d43-8505 VEM 901 0602 160591.30
0002-3d43-8506 VEM 901 0702 160230.70

3. “show system redundancy status” shows operational mode “none”, which should be HA:

vsm# sh sys red stat
Redundancy role
---------------
administrative: primary
operational: primary


Redundancy mode
---------------
administrative: HA
operational: None


This supervisor (sup-1)
-----------------------
Redundancy state: Active
Supervisor state: Active
Internal state: Active with warm standby


Other supervisor (sup-2)
------------------------
Redundancy state: Standby
Supervisor state: HA standby
Internal state: HA standby