New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
I disagree.
With what statement? For now, I'm having troubles with all their software.
If you are really concerned with PPS and DoS - even at just dual 1 Gbit ports, you should make the investment on something more realistic for the hosting environment. If its just for a business (single business) thats a different story. A juniper MX240 base minimal package should be had off ebay/second market around $10k. Yes thats 3 times your $3500 budget, but if you get the vendor to finance on 3 year, and your a hosting company, this should be a non brainer and give you a lot more across the board.
A 3BXL, when you are just pushing 1-2 Gbit would work fine with a DFC.
However, the 3BLX that we have been using seems to choke every time it pulls full tables from two peers when there is 6-7gb of traffic going over it at about 2.5 Mpps.
Honestly would sink the money into a MX240 or equivalent. It will future proof yourself for a good deal of time.
I've deployed many of there products in many real world scenarios, including ERL, switches, and cameras, and any problems have been promptly resolved with updates after contacting Ubiquiti by posting on their forum.
The 3BXL is still just fine for 2-3 fulls and 10gbps. You need to make sure your config is right, you have configured for fast table loads, and you are on a stable 12.x release. 15.x is so full of memory issues related to BGP it's not worth your time.
Yes, bgp convergence time on 6500 w/3bxl is slow... but it's always been slow.
The super big negative thing about them is they are power hogs.
Sounds like you have something seriously wrong with your setup, including possibly mis-matched or cheap modules (X6100 series, etc).
A 6500 series with SUP720-3BXL will run a ton of bandwidth, a ton of PPS, and a ton of arp (for machines behind it).
You can run 10Gbps+ with 4+ tables and be completely fine.
You can run 4 full tables on 1 Gb of ram?
The amount of memory a full table takes up is fairly miniscule, actually.
On a properly engineered solution (note that this is the RIB, not the FIB), it's not uncommon to see > 10 full tables within a gig of mem.
OpenBGPd is one such example of an extremely well engineered solution.
I however am not too sure of antiquated Cisco solutions, they aren't exactly known for being optimized..
Do you even realize how many hosting companies are run on Cisco 6509 with SUP720-3BXL as their cores? or edges? or their collapsed core+edges?
Do you even realize WHO those hosts are? Let me give you a hint, its a huge part of the industry and everyone seems to be doing just fine for the most part.
Power usage is a beast on those, but by the amount of people loading up eBay shit servers (L5420, L5520, etc) and selling them for almost free -- it seems like a lot of people have found free electricity.
That does not do much to show how that is done.
Like how you gona run 4+ FullView tables on SUP720-3bxl if it's has only 1000000 TCAM space? FullView for now is more than 512k. And dont forget about ipv6.
--Removed
You guys are just jumping on each others forum peepee's right now.
Like I said, it works fine -- it can hold 1MM ACTIVE routes in FIB and then the rest is in software.
I dont use Cisco 6509's so its not exactly like I have any reason to care on what you guys believe or not...
I'm running Juniper MX's for edge/border and I am not collapsing edge/core either.
Like I said though, the 6509 w/ sup720-3bxl is what a ton of hosting providers are using -- they are managing to keep your servers and vps's online some how with all those bgp sessions, right??
MUST BE MAGIC AGAIN. There's a lot of "magic" that apparently happens on LET.
6500's are great legacy switches but for routing its just not going to be comparable to Juniper or Cisco's modern routers.
Take one of those 3BXLs stick 2 full sets of BGP routes and then turn on netflows
We don't use any of them anymore but the 6500 series is definitely the platform that built the modern hosting industry.
Do you even understand how BGP works?
Your localized table is not gong to be 2mm routes if you are pulling in 4 x Full Tables.
Yeah, my bad, it was like 6:00am, so I say something stupid.
Of course, there are rib and fib. Rib can hanle as mush fullviews as your ram can handle.
6509s are workhorses, but I think there are wiser ways to spend money these days. Plus, you need a nuclear reactor nearby to power them.
I wish I knew what half of this stuff was. Obviously it's routing technology, but is there any place you guys would recommend checking out to learn about this fascinating field?
There actually is really not anything available in the same price range as 6509 w/ sup720-3bxl that has same performance capabilities.
You will be spending a lot more.
The 6509E w/ SUP720-3BXL is still a really good choice for distribution/core layers. It has a massive ARP table and will be great for that layer, but then you would still need an aggregation layer for 10G ports though as 10G density on the 6500 series is low.
Adding 10G aggregation switches are like $5K-$7K per.
I wouldn't run it on the edge/border doing eBGP however. You just need to find the money to get better gear on that layer ($25K - $50K PER router on edge/border).
Thanks for all the great replies, I got stuck on a huge project and now just getting back to this. I am not co-located anywhere. We are a very small, self-contained data center and we host just a couple-hundred customers and about 75 servers. We have our own AS and do BGP, and I prefer full routes. I could filter out /24s I guess but I'd rather not. We are a purely managed hosting company for our IT Consulting customers. For high bandwidth stuff, we farm that out and manage it for our customers.
We have 2 Gig ports, but they are typically less than 80Mb/sec. They are burstable with 100Mb commits. Most of our stuff is not high bandwidth.
We do have a TopLayer (don't ask, we got it free) in the mix that blocks a lot of stuff before it hits the Vyatta router. Vyatta running on an Sun Sparc Dual Quad Core. We are using the Subscription edition of Vyatta, our 3 year maintenance is up. Brocade raised the price considerably. I don't like the company, at all.
The TopLayer is pretty much on its way out too and I want a router that can handle some bigger attacks. I can't afford the latest and greatest.
I think thats why I was so attracted to this forum, low end talk. People always tell me I can't do this or that, but somehow I manage to make it work. Not everyone can be Rackspace or Amazon.
Within 2 years, I don't think I'll be hosting anything in-house. It's getting cost prohibitive to do so and the big providers make it more attractive all the time.
So I just need something to get me by for the next couple years. I was looking at this, which seems to have some redundancy:
http://www.ebay.com/itm/371177691471
I also looked at those EdgeRouters. They almost seem TOO cheap? Aren't they just running Vyatta on them?
Considering a 650X does 10G for not even 15kEUR the 2kW power usage are negligible for many.
Single supervisor and you'll want to check costs of linecards for that unit. There are cheaper 6509's floating around and you can probably pick up a Juniper M20 for similar money with some interfaces.
Line Card? You mean like a 48port GIG Switch? They are under $100. I'd probably buy another supervisor in a few months so there's 2.
Also the 6509's use more power, like 3000+ watts. Kinda crazy. I do have cheap power, but damn.
There are multiple models of gigabit card, some will be compatible with 3BXL some wont. Also I think you'll need to upgrade the PSUs if you want to run dual 3BXL's on that chassis. Though can't say I have tested dual 3BXL on anything other than 6509 and 6513.
From memory (check this, its been years since I went near Cisco 6500), you will need WS-X6748-GE-TX for a compatible line card.
WS-X6748-GE-TX are < $200 on eBay. I am still looking at $4,000 for a sinking boat anchor. It's crazy what aftermarket Cisco goes for. This is 10+ year old technology.
Maybe I should just stick with my Vyatta.
I think all the solutions like 6k, M20 and so on isn't the right solution here. This can be done from a VM to a small router like CRR1009 or ERpro by this amount of traffic. Why you should spent so much money, when you can do this with two small devices for under <1k and you will have a redundant setup which can fully fulfill 200 Mbit/s without any problem?
This make no sense and burn many amount of energy and money.
That's a very bad idea!