Re: [j-nsp] Advice on a 100Gbps+ environment
Arista is still the best deal around when it comes to very high speed, high density ethernet. In some deployments, it's the only possible choice. Juniper doesn't have a great product offering at ToR and even access layer/core routing when you start talking 40 gig ports. On Tue, Jul 2, 2013 at 11:00 AM, Morgan McLean wrx...@gmail.com wrote: Wow, this thread snowballed into quite the MX80 debate. For the record, I run two in production where I am employed full time and they perform beautifully, though woefully underutilized. Using static routes and /32's as peering endpoints is a great option I skimmed over, I'll see if the upstream can do this...they should. Unfortunately, the customer signed the contract for bandwidth with inteliquent; we have existing 10G with telia and 10G with cogent along with a couple existing 10G from inteliquent, but I'm not sure if they'll stay. So I didn't really have much say...I think the price point was more important than the benefits of signing to a few carriers. In short, I'm working on that. This traffic should be mostly web. Sorry, I meant to say OSPF and ECMP. I would like to be able to run the VRRP at the end of row and extend L3 as far as I can, but I guess the customer wants to be able to spread machines in the same environments among multiple rows, which is understandable, but that means I need to run L2 from distribution to access. Each row needs 100gbps useable, so I suppose 4 x 40GBE LAGs would do the trick nicely. If my client doesn't want to spend the money in that area... Any good aggregation switch suggestions? Juniper is doesn't provide good ports for $ in the switching realmcustomer balked at the cost for a four port 40G blade on a 9200. Might check out brocade.. Thanks, Morgan On Tue, Jul 2, 2013 at 8:13 AM, Christian de Balorre cdebalo...@neotelecoms.com wrote: Slow control-plane. No RE redundancy. More limited rib fib than regular MX. Cryptic licensing scheme. Otherwise nothing really wrong. Christian Le 02/07/2013 15:55, Drew Weaver a écrit : And what is wrong with the MX80 as a peering/transit router for up to 80Gbps of traffic? Thanks, -Drew -Original Message- From: juniper-nsp [mailto:juniper-nsp-bounces@**puck.nether.net juniper-nsp-boun...@puck.nether.net] On Behalf Of Dobbins, Roland Sent: Tuesday, July 02, 2013 9:01 AM To: juniper-nsp Puck Subject: Re: [j-nsp] Advice on a 100Gbps+ environment On Jul 2, 2013, at 7:19 PM, Mark Tinka wrote: Says who? Doh - MX*480*, not MX*80*. My mistake. --**--** --- Roland Dobbins rdobb...@arbor.net // http://www.arbornetworks.com Luck is the residue of opportunity and design. -- John Milton __**_ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/**mailman/listinfo/juniper-nsp https://puck.nether.net/mailman/listinfo/juniper-nsp __**_ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/**mailman/listinfo/juniper-nsp https://puck.nether.net/mailman/listinfo/juniper-nsp __**_ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/**mailman/listinfo/juniper-nsp https://puck.nether.net/mailman/listinfo/juniper-nsp -- Thanks, Morgan ___ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp ___ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp
Re: [j-nsp] Advice on a 100Gbps+ environment
The QFX3600 is probably a little expensive for L2, but 64x 10GE ports in a 1RU ToR (or 16x 40GE, or a combination in the middle) is pretty solid. According to the docs they also support link aggregation up to 32 members[1]. Would be interesting to know if this allows 40GE ports to be used natively in an aggregated ethernet... [1] http://www.juniper.net/techpubs/en_US/junos12.3/topics/reference/general/qfx-series-software-features-overview.html#high-availability-features-by-platform-table On Tue, Jul 2, 2013 at 11:00 AM, Morgan McLean wrx...@gmail.com wrote: Wow, this thread snowballed into quite the MX80 debate. For the record, I run two in production where I am employed full time and they perform beautifully, though woefully underutilized. Using static routes and /32's as peering endpoints is a great option I skimmed over, I'll see if the upstream can do this...they should. Unfortunately, the customer signed the contract for bandwidth with inteliquent; we have existing 10G with telia and 10G with cogent along with a couple existing 10G from inteliquent, but I'm not sure if they'll stay. So I didn't really have much say...I think the price point was more important than the benefits of signing to a few carriers. In short, I'm working on that. This traffic should be mostly web. Sorry, I meant to say OSPF and ECMP. I would like to be able to run the VRRP at the end of row and extend L3 as far as I can, but I guess the customer wants to be able to spread machines in the same environments among multiple rows, which is understandable, but that means I need to run L2 from distribution to access. Each row needs 100gbps useable, so I suppose 4 x 40GBE LAGs would do the trick nicely. If my client doesn't want to spend the money in that area... Any good aggregation switch suggestions? Juniper is doesn't provide good ports for $ in the switching realmcustomer balked at the cost for a four port 40G blade on a 9200. Might check out brocade.. Thanks, Morgan On Tue, Jul 2, 2013 at 8:13 AM, Christian de Balorre cdebalo...@neotelecoms.com wrote: Slow control-plane. No RE redundancy. More limited rib fib than regular MX. Cryptic licensing scheme. Otherwise nothing really wrong. Christian Le 02/07/2013 15:55, Drew Weaver a écrit : And what is wrong with the MX80 as a peering/transit router for up to 80Gbps of traffic? Thanks, -Drew -Original Message- From: juniper-nsp [mailto:juniper-nsp-bounces@**puck.nether.net juniper-nsp-boun...@puck.nether.net] On Behalf Of Dobbins, Roland Sent: Tuesday, July 02, 2013 9:01 AM To: juniper-nsp Puck Subject: Re: [j-nsp] Advice on a 100Gbps+ environment On Jul 2, 2013, at 7:19 PM, Mark Tinka wrote: Says who? Doh - MX*480*, not MX*80*. My mistake. --**--** --- Roland Dobbins rdobb...@arbor.net // http://www.arbornetworks.com Luck is the residue of opportunity and design. -- John Milton __**_ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/**mailman/listinfo/juniper-nsp https://puck.nether.net/mailman/listinfo/juniper-nsp __**_ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/**mailman/listinfo/juniper-nsp https://puck.nether.net/mailman/listinfo/juniper-nsp __**_ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/**mailman/listinfo/juniper-nsp https://puck.nether.net/mailman/listinfo/juniper-nsp -- Thanks, Morgan ___ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp ___ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp ___ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp
Re: [j-nsp] RIB and FIB - Memory for MX with LR
Yes L3VPN services are the biggest glutton of RIBs and FIBs, and already exceed several millions of VPNv4 prefixes for some of the ISPs couple years ago. That's why they have been using several distinct route-reflector planes and hierarchies to be able to scale the distribution of intra and inter AS prefixes as well as service dedicated PEs in major POPs to alleviate the edge constrains i.e. number of VRFs/FIB entries per PE/Line-card. It kind of makes me sad to see these state of art designs being replaced by several boxes capable of holding 22Megs of prefixes in RIB. adam ___ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp