I just ran a test across an AirFiber link with 384 byte packets. The interface shows it passing about 650Mbps going from an i7 x86 to a CCR with existing Internet traffic.
On Fri, Jan 24, 2014 at 8:20 PM, Tom DeReggi <[email protected]>wrote: > Sam, > Thats quite impressive, to be able to support that many queues and filter > rules. So apparently, those key services must be multi processor. > That is good to learn. > > Eric, > Regarding single core apps. It may not matter all that much if an app is > single core, if it can use a unique core. > My concern is if key single core apps default to sharing the same core. > > Faisal, > A single 1.2G processor per port is probably fine for large packets and > Full throughput. Im concerned on whether a single 1.2G core would be enough > for full throughput with average small packet sizes or DDOS situations. > With X86 processors, in the past we've shown it was not. But then again, > the CCRs arent X86, and our past 4core X86 test machines, didnt have 36 > procs to handle the load of other processes. > > Paul, > Since we are on a budget, and need something to put in place quickly w/ > SPFs, sounds like the 36core CCRs will solve our immediate need for Core > BGP Router. It clearly will do way much better than the 1100 dual core > that we temporarilly put in place, until we had time to order in a CCR. > > Whether the CCR will handle our growth plans for head end, thats yet to be > seen. > In our application we wont have nearly the number of rules per router as > Sam's example, as we do filtering and bandwidth management at each tower, > to spread out the load. > > Last Question: > Long term, what Im most concerned about is how much throughput can be > passed per gig port. Meaning how close to theoretical wirespeed. Because > when calculating a providers cost per MB, its a big difference whether a > router port can push the full GB versus say 50%. > It can double a provider's cost per MB, requiring duplicating ones fiber > infrastructure prematurely. > > Has anyone tested how small the average packet size can be and still > achieve theoretical wirespeed, in a simplified configuration over a single > port? > > 1Gbps FDX, can 90% of that be acheived with 384k avg packet size? > > > > Tom DeReggi > RapidDSL & Wireless, Inc > IntAirNet- Fixed Wireless Broadband > > > > ----- Original Message ----- > *From:* Sam Tetherow <[email protected]> > *To:* WISPA General List <[email protected]> > *Sent:* Friday, January 24, 2014 5:28 PM > *Subject:* [Spam] Re: [WISPA] Mikrotik on Multi-core > > Replaced an aging powerrouter 732 with a CCR-1036. Set up as a > transparent bridge for traffic shaping. Passing 478M peak with 8200 > interface bridge filter rules and 8000 queue tree entries, cpu utilization > peaks at about 50 and all 36 CPUs are in use according to /system resource > cpu print > > The 732 started giving us CPU limitations at about 240Mbps. The whole > thing could be reworked so we didn't have so many filter rules or queue > tree entries, but the original installation replaced a MAC based bandwidth > limiter and they wanted to keep that setup. > > Other than some lockup issues we had on ROS versions before 6.7 we have > been pretty happy with the box and for under $1k it is hard to beat. > > > On 01/24/2014 03:53 PM, Tom DeReggi wrote: > > Hi everyone. Been awhile since Ive been here, so not sure if this is a > redundant topic or not. > > Anyone got experience with Mikrotik on their newer Multi-Core platform, > using as a Core Router for interconnecting multiple Gig backbone > connections (w/ BGP, OSPF, Queues, Firewalls, VLAN tagging)? > > To be more specific.... Comparing Mikrotik's 36 core 1.2Ghz models to > say a third party Quad core 3Ghz model. > > What do we need 36 cores for, when we got 11 eth ports? Are they even used > by software? Is later Mikrotik Firmware allowing.... > - multiple processors to handle a singe NIC port? > - which Mikrotik software features are able to effectively spread > accross to a unique processor or use multiple processors? > > Is 1.2Ghz enough? > > Do the embedded NICs in the 36core units pass full Gig capacity? (In past, > we learned depending on which NIC and driver brand, a NIC could pass as low > as only 30% of full capacity w/ large packets, where as a later generation > PCIE w/ ATIO Intel could pass upward of 90% of full capacity w/ small > packets.) > > Im asking because back in the day, there were many Linux services relating > to routing that were written to be only single processor support. > Because of this, it was important to have the highest speed Ghz processor > possible, since some critical services (the bottleneck) would share only 1 > primary processor, regardless of how many processors were in the router. > > In past experience specific to Mikrotiktik, I often ran into issues with > added features (firewall rules, Queues, etc) drastically draining the > processing power of a MT router slowing throughput way below the > theoretical published port throughput. > > For example, can Queues or Firewalls spread multiple processors? > > Can these 36core units handle bandwdith management (Limiting or Queues) > for high speed subscribers, such as 100mb and 200 mbps customers? > In the GUI of v6.7, I dont see anything higher than 2mb or 10mb depending > on location of parameter. > > > > > > Tom DeReggi > RapidDSL & Wireless, Inc > 301-515-7774 > IntAirNet - Fixed Wireless Broadband > > > _______________________________________________ > Wireless mailing > [email protected]http://lists.wispa.org/mailman/listinfo/wireless > > > ------------------------------ > > _______________________________________________ > Wireless mailing list > [email protected] > http://lists.wispa.org/mailman/listinfo/wireless > > > _______________________________________________ > Wireless mailing list > [email protected] > http://lists.wispa.org/mailman/listinfo/wireless > >
_______________________________________________ Wireless mailing list [email protected] http://lists.wispa.org/mailman/listinfo/wireless
