Hi All,
I am not sure is this the good place to ask this question or not!!!
I am looking for feed back on having OSPF multi-area, lets say if you have
multiple location in nonbackbone areas and those nonbackbone areas are
connected with the one backbone area. For example: OSPF AREA1 has the
connectivity to OSPF AREA0 using two ABR, so what is the optimum way to
achieve the load balancing or load sharing for traffic entering or leaving
the area, what are the possible way to configure it?
regards
Devang Patel
First, without any details, it sounds like you might be better off with static routes than with OSPF. I'm not trying to be patronizing, but you don't mention many details, and some of the details you omit are the crucial ones for OSPF.
1. Do these remote areas have multiple paths to the central area for failover? E.g. a 10Mbps Metro Ethernet primary link, and a 1.5Mbps DSL secondary?
2. Does the central area have multiple routers for failover? E.g. a Cisco 7200 for the incoming Metro Ethernet primary connections, and a Cisco 3660 for the slower secondary connections?
3. Are there any tie-ins between the remote sites that bypass the central site? E.g. site1 and site2 both communicate to the central site via Metro Ethernet, and they also communicate to eachother via DSL.
If none of the above are true, then static routes would be better for you (for the remote area/s in question). E.g. area1 has multiple paths, so ospf is warranted; however, area2 has just one path so a static approach would usually be better.
Your language seems to indicate that OSPF is warranted (area0, area1, two ABRs). I am assuming you mean Area Border Router not Associative Based Routing (vs. OSPF). I am also assuming this is a non-public system (internal network, probably a MAN or WAN).
If so, without any further details, I would set it up for bandwidth/failover. Weight the paths appropriately. Keep it as simple as you can. OSPF can become a morass.
If you sketch your situation out more, we can be more helpful.... Campus? MAN? How public? Multi-pathed? Multi-homed? Multiple interlinks? Are there some lines with reliability problems where the lower bandwidth links are actually preferred? Do you have any decentralized concentration points that might have problems due to multiple remote sites shuttling traffic through it (due to multiple interlinks)?
--p
devang patel wrote:
Sorry about that!!!
1. Do these remote areas have multiple paths to the central area for
failover? E.g. a 10Mbps Metro Ethernet primary link, and a 1.5Mbps DSL
secondary?
2. Does the central area have multiple routers for failover? E.g. a Cisco
7200 for the incoming Metro Ethernet primary connections, and a Cisco 3660
for the slower secondary connections?
3. Are there any tie-ins between the remote sites that bypass the central
site? E.g. site1 and site2 both communicate to the central site via Metro
Ethernet, and they also communicate to eachother via DSL.
Answers:
I have two T1 line to the non-backbone area and both T1s are terminated to
the two different routers on non-backbone area as well as to backbone area,
and I dont want to achieve primary and secondary role, I want to go for the
load sharing kind of scenario. All sites are connected with the central
site.
ABR means Area border router only.
I am attaching one generalized diagram, please look at that one.
Now I want to achieve the load balancing between the traffic going from R1
to R8, I want to achieve some of the networks on R1 should be reachable via
R2 and some of them via R3 for the traffic coming from the R8. assume all
links are same.
regards
Devang Patel
Patel,
I would suggest you to read a few things about the path selection algoritm....as if i understand your words you are asking for an issue on LSA type 4 rather than multiple AS and therefore LSA type 5 /7-ASBR
prefer backbone intra-area paths over inter-area paths....
Excerpted from RFC 16.4.1...- When multiple intra-AS paths are available to
ASBRs/forwarding addresses some rules using different costs apply when the same ASBR is reachable through multiple areas, or when trying to decide which of several AS-external-LSAs should be preferred. In the former case the paths all terminate at the same ASBR, while in the latter the paths terminate at separate ASBRs/forwarding addresses.
http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a0080124c7d.shtml
.//ID
It is my understanding of OSPF that, if you have paths with equal distance and cost to a destination, load balancing happens automatically for up to four (or is it 6 for OSPF?) clear paths. In your diagram R1 to R8 load balancing should happen naturally, unless you have weighted one of the paths. You have much more than 4 paths here, so you should weight the ones you want. E.g. 1-2-4-6-8, 1-3-5-7-8 would be the most straightforward, and barring some type of natural concentration of bandwidth (e.g. R3 having 10X the hosts connected that R2 has) it would be the easiest to implement. This only applies to coequal routing (e.g. all links are T1s). If you are doing unequal routing I think you are out of luck. I would stick to two paths if possible for simplicity's sake. OSPF can become a quagmire if you let it.
So, first step is weight your chosen paths equally, and make sure they are preferred over other possible paths.
Second step is to decide what kind of load balancing you want: per packet, or endpoint. If you set it up per packet, you get equal load balancing with the chance of out-of-order packets on the other end. It can also take up a lot of the router's cpu resources. If you decide on endpoint load balancing you will almost always have one path taking the majority of the traffic--e.g. all traffic to the file sharer will take path1 and all traffic to the ntp server will take path2, and path1 will definitely be more heavily loaded. To properly balance by endpoint takes some micromanagement.
Depending on your router, you turn ip route cache on for endpoint balancing, and turn it off to enable per packet balancing.
Cisco has something called CEF which I have never used, which supposedly enhances OSPF load balancing--uses special algorithms to speed it up.
--p
OSPF (on cisco anyway) will balance 6 paths automatically assuming you
haven't messed with bandwidth or cost settings, and the paths have the
same iftype. If not, setting the bandwidth equally will also do the
trick. (I don't like messing with cost directly, just me)
I would also point out that you would probably be better off not using a
multi-area config for this solution. My rule of thumb, which may be a
bit antiquated now, is create multiple areas when you have either more
than 2000 routes, or more than 300 interfaces total, or 200 on a single
router within a single area. This was just a general rule of thumb
based on some problems with certain hardware scaling past that point and
behaviors during reconvergence. If you can use a single area, go for
it. If you are the type of company that might get acquired someday, or
hopes to, you can save yourself time and select an AREA ID that is based
on address space you own, that way there will be no AREA ID overlaps and
you would have the possibility of connecting to someone else's area 0
(acquiring company).
Also, don't use per packet load balancing. It may work in the short
term, but it almost always screws with an applications performance.
My $0.02.
Steve
