Does anyone have any success with the Brocade SLX 9540 or similar? Its going to be taking full BGP tables from two Tier1's and some peering.
The specs and sales rep says its fine, but the price makes me think its too good to be true.
We are trying to shepherd an old Cat 6509 out of our core.
Kevin Burke
802-540-0979
Burlington Telecom - City of Burlington
200 Church St, Burlington, VT 05401
It won't hold a full table. 256,000 IPv4 and 64,000 IPv6 routes.
Last I heard (before switching shops), not yet it won’t.
Best regards.
That won’t hold a full table - so performance isn’t relevant.
-Ben
That was changed earlier this year AFAIK. The website was slow to get updated but has been updated now. Current claim is 1.5M IPv4 and 140k IPv6. You need the "advanced feature license" to get access to that.
140K IPv6 equates to about 560K IPv4 routes, leaving the end user with 940K IPv4, which is not a lot of ceiling space considering we're at 741K IPv4 + and 60K IPv6 (240k IPv4 equivalent) now (941K total). This will leave you with 559K. I am not sure what the OP has for peering but with trying to keep 20% of TCAM space free, and keeping up with the current rate of rise according to CIDR-report, I'd say 4 years product lifetime if the OS has excellent TCAM management.
Considering how the device looks like a switch and the SLX9850 uses Broadcom sillicon, I'm thinking it must use the Jericho chipset or some variant to get that kind of performance. In the end, your mileage may vary.
If you buy brocade, be sure to also by a license for securecrt so that backspace works over ssh…
also, just don’t do brocade… ever.
+1 SecureCRT in general, and don’t buy Brocade,
I was happy when I got to pull out the last Foundry.
140K IPv6 equates to about 560K IPv4 routes, leaving the end user with 940K IPv4, which is not a lot of ceiling space considering we're at 741K IPv4 + and 60K IPv6 (240k IPv4 equivalent) now (941K total). This will leave you with 559K. I am not sure what the OP has for peering but with trying to keep 20% of TCAM space free, and keeping up with the current rate of rise according to CIDR-report, I'd say 4 years product lifetime if the OS has excellent TCAM management.
I'm actually in the process of spec'ing one of these (if indeed it's appropriate) for a limited full-Internet-routes application and indeed these are the questions I've been asking of my rep.
On "classic Netiron" (MLX, etc.) the numbers they often quoted were actually somewhat pessimistic in that they were one of their stock TCAM profiles, and you actually ended up with BOTH the IPv4 and IPv6 route counts simultaneously.
Considering how the device looks like a switch and the SLX9850 uses Broadcom sillicon, I'm thinking it must use the Jericho chipset or some variant to get that kind of performance. In the end, your mileage may vary.
I want to say it's a Qumran. They apparently have a bigger SLX pizzabox in the works that claims 4M IPv4 FIB and some stupid amount of buffering (8GB IIRC?). I know that's a Qumran, but that also seems like a truly huge amount of TCAM, so I dunno if that's with "typical aggregation" or some other shady trick.
Works fine for me using OpenSSH in most Linux-y terminal emulators (Konsole, Linux console, Gnome terminal). I didn't do any special configuration.
Now, over serial, enjoy your ctrl-H unless you do some remapping.
I've never had any real problems with the hardware. The software can leave something to be desired especially on the old Foundry stuff that can't run the modern software, but if you just want it to push packets all day long, they seem to be pretty stable.
Only bug I've been bitten with recently is apparent CAM corruption when manipulating large ACLs, but that was on the old (EOL) FCX platform. It's stable as long as you don't CHANGE things, and networks never change, right? (/s)
Netiron seems to be more stable but definitely lacking control plane features, especially for MPLS, including some major ones that I gather the big C and J have had available for quite a while.
I'm curious how things will diverge now that the "switching" line is at Ruckus/Arris while the "routing" line is at Extreme. I can't say I was ever a fan of Extreme's software, either, and I don't really have enough experience with Arris gear to comment.
Certainly like any vendor's box, know what you're getting. It's a packet pusher.
On a similar/related topic, has anyone used the Juniper MX204? It seems to occupy roughly the same space as the SLX9540. Less bandwidth, but JunOS is presumably more fully featured in terms of Internet-scale stuff one might want.
Yep, they fixed backspace via SSH (at least for MLX) a few years ago.
Sad that they didn't fix the console ports at the same time.
I’m just going to echo what a few others have been saying. Brocade (now Extreme) have come a long way since the Foundry days; and the SLX isn’t based on the old Netiron code. The platform is a completely different animal.
I’ve been a happy Brocade customer for a while now.
Hi,
I do have some 9540s near exchange points, but they are not 100% productive right now, basically waiting for the next software release this month and a maintenance window. In my eyes the device is filling the gap between the CES/CER series and the MLX/SLX9850. It will be also interesting where Bro<H><H><H> Extreme is going to position the new, bigger (?) brother 9640 next year.
Our 9540s take full feeds right now without moaning:
show ip route summary
IP Routing Table - 717510 entries:
Number of prefixes:
/0: 1 /8: 14 /9: 11 /10: 35 /11: 98 /12: 291 /13: 567 /14: 1130 /15: 1932 /16: 13357 /17: 7876 /18: 13735 /19: 25102 /20: 38379 /21: 45371 /22: 89251 /23: 73451 /24: 406788 /25: 2 /26: 9 /27: 22 /28: 35 /29: 36 /30: 8 /32: 9
And for this to work, you need
- the latest release, 18r.1.00, better 18r.2.00 (End of Nov 18) because of DEFECT000666685 (prefix filter list)
- buy / activate the trust-based advanced features license for MPLS, BGP-EVPN, CE2.0, Optiscale, basically on cli:
=> license eula accept
- activate Optiscale Routing in the configuration
=> profile route route-enhance hw_opt on v4_fib_comp on v6_fib_comp on
=> shall scale to 1.5M IPv4 & 140k IPv6 routes (the 256k IPv4 / 64k IPv6 information is old)
With the advanced license you are also eligible to spin up a Linux VM for additional tools, monitoring et al on a reserved cpu core.
The software part of the SLX feels like a mixture of Brocade NOS (port-channel, vlan, switchport, … ) and Netiron components (router configuration, …) while the hardware is pretty much inspired by the VDX line.
Therefore some SNMP tools like LibreNMS recognize the SLX as VDX because of fancy wildcards in the hardware model type and I am still looking to write some patches for this.
Regards
Jörg
Thanks for everyone who responded on and off list.
As a small company that is happy to still be in business the pricing is too good to ignore. A “gently used” ASR-9006 is something like $45k for one plus a shelf spare. A brand new SLX 9540 is something like $30k for one plus a shelf spare.
There were some common things. Software is behind where we would like. The occasional bug like that SSH one. Also there are some relatively common features like IPv6 outbound ACL and BGP MED that aren’t there. This stuff isn’t a showstopper but I will take this a sign of things to come.
As for the notes about full tables. Different vendors seem to have used different techniques to get past the hard FIB limit that we are all used to. I had the same question when pawing through the spec sheets. So I asked the sales rep:
“We can support 1.5M routes……
These platforms support all of the requirements detailed above for Internet routing. In particular, they support a table size of 1.5 million IP routes today, ensuring headroom for the next 5-7 years. This scale is made possible through our new technology called Extreme OptiScale™ for Internet Routing that optimizes programmable hardware and software capabilities to accelerate innovation and deliver investment protection.
https://www.extremenetworks.com/extreme-networks-blog/internet-routing-in-the-enterprise/”
Kevin Burke
802-540-0979
Burlington Telecom - City of Burlington
200 Church St, Burlington, VT 05401
I think Extreme is doing the same thing with their Extreme OptiScale™ that Arista is doing with their Arista FlexRoute™ and EOS NetDB™. They are both using Broadcom Jericho /Qurman with extenal TCAM, but still has a hardware limitiation on route table size. Then in software they filer right?
Question is who has a better solution Arista or Extreme for this?
Also, the question is can any whitebox vendors do the same thing, with the same Broadcom switch you can buy for around $9k new.
Another question, could you even consider these with the Juniper MX204 coming in at $20k?
Hey,
They all do in principle the same thing. There are memories for
longest path lookup and memories for exact lookup. I believe the trick
is to put specific prefix size, like /24 to exact lookup table,
relieving the LPM table stress greatly. Then in parallel ask both, and
take more specific result.
There are variation to this, like having multiple separate exact match
tables, and populating each with different prefix size, and so forth.
Juniper on PTX is doing something quite different, they are asking
on-chip bloom filter about hint on where to query, reducing query
count they need to do towards high latency off-chip memory.
MX is doing yet something different, having JNPR proprietary memory
ASIC (no longer plain (RL)DRAM).
ASR9k is still just TCAM (for all ezchip generations, unsure about lightspeed).
Nicolas Fevrier has a very detailed blog post on how Cisco handles the prefixes on their Broadcom Jericho based NCS 5500 gear.
https://xrdocs.io/cloud-scale-networking/tutorials/2017-08-03-understanding-ncs5500-resources-s01e02/
I'm pretty sure the principle is more or less the same for the Jericho based platforms on Arista and Extreme.
Best regards,
Chris
Some of it is Extreme, some of it is Arris.
The only issue I’ve had with anything Brocade\Foundry is lack of features in older platforms. They’ve always been solid for me.
I love the nitty gritty detail in this author's post and I'm glad he concludes by stating clearly that while the base card (spec sheet says: "On-chip tables for 256K IPv4 or 64K IPv6 routes" and "On-chip tables for 786K IPv4 host routes, MAC, and labels") can actually hold a full BGP table today when configured appropriately, Cisco still recommends the scale cards for that application (spec sheet says: "FIB scale up 2M IPv4 or 512K IPv6 routes" and "On-chip tables for 786K IPv4 host routes, MAC, and labels").
I do have to wonder about the internal expansion of each /23 route into two /24 routes in their FIB algorithm, as I would have thought Cisco would have attempted to go the opposite way, but I'm sure Cisco has their reasons.
I have no horse in this race, however one need only look at the NYIIX
outages list to see how well the Brocade/Extreme SLX platform works on
at-scale service provider networks...