RE: perf #s for GRF vs 7500 Re: Anyone Deployed Ascend's GRF IP S witch?
Rob, Comparing your BAY routers to everyone elses is like comparing a top fuel dragster to someones 69 Chevelle street racer. You folks have some heavily modified software and hardware, from what I understand. However, we've got a couple of BAY BCN/BLN routers about, and the numbers that have been mentioned previously in this thread are fairly accurate. Chris ---------- From: Rob Skrobola[SMTP:rjs@ans.net] Sent: Wednesday, August 27, 1997 12:02 PM To: Tony Li Cc: Paul Peterson; nanog@merit.edu Subject: Re: perf #s for GRF vs 7500 Re: Anyone Deployed Ascend's GRF IP S witch? Folks, We have bcn/bln's out there with over 60 bgp peers on a 64Mb ARE. Works fine. Taking in about 63000 pps (170 Mbps) over 6 interfaces with a high of 20k pps when I looked a couple of minutes ago..Not untypical of the 30 bcn's and bln's on our network.. So the 4-6 Mb per peer thing is inaccurate. On the way high side. RobS BGP Peers --------- Local Remote Remote Peer Connection BGP Total Address/Port Address/Port AS Mode State Ver Routes --------------------- --------------------- ------ ------- ---------- --- ------ .. 64 peers configured. Memory Usage Statistics (Megabytes): ------------------------------------ Slot Total Used Free %Free ---- -------- -------- -------- ----- 6 61.67 M 32.82 M 28.84 M 46 %
Subject: Re: perf #s for GRF vs 7500 Re: Anyone Deployed Ascend's GRF IP S witch? From: Tony Li <tli@juniper.net>
paulp@winterlan.com (Paul Peterson) writes:
Bay claims to hold the entire Internet routing table in just 4-6MB RAM per BGP peer (I assume this is after convergence). They say that the method in which they do this is proprietary. I am just wondering if it is possible.....
That's certainly possible. However, it would be interesting to see how it scales with the number of peers. You could quickly find yourself needing
64MB if it's even just linear.
Tony
Chris, Really, I wasn't comparing. Mostly I was just responding to the bgp memory usage issue, and noted load on the boxes to indicate that these were not unused boxes doing that number of peers.
From: "Chris A. Icide" <chris@nap.net>
Comparing your BAY routers to everyone elses is like comparing a top fuel dragster to someones 69 Chevelle street racer. You folks have some heavily modified software and hardware, from what I understand.
Actually, we use stock bcn/bln hardware. As to general availability of the software mods, I don't really know.. RobS
However, we've got a couple of BAY BCN/BLN routers about, and the numbers that have been mentioned previously in this thread are fairly accurate.
Comparing your BAY routers to everyone elses is like comparing a top fuel dragster to someones 69 Chevelle street racer. You folks have some heavily modified software and hardware, from what I understand.
Not exactly. The hardware is stock and most of the software (atleast as of a couple months ago) mods can be found in released code. Jeff
On Wed, 27 Aug 1997, Chris A. Icide wrote:
Rob,
Comparing your BAY routers to everyone elses is like comparing a top fuel dragster to someones 69 Chevelle street racer. You folks have some heavily modified software and hardware, from what I understand.
AFAIK, ANS has stock hardware (FRE/ARE, with ARE performing soloist duties), same as us and other Bay cutsomers. Software is another issue though. Rob can probably shed some extra light on this, but i'll add what i've gathered. Their code line supposedly split at 10.x, while mainstream progressed up to 11.x (w/ISP mode having *some* of the ANS addendums). 12.x is supposed to bring the code lines closer together, though i'd imagine ANS will still have needs that wont be rolled into the mainstream code.
From: Rob Skrobola[SMTP:rjs@ans.net]
Folks, We have bcn/bln's out there with over 60 bgp peers on a 64Mb ARE. Works fine. Taking in about 63000 pps (170 Mbps) over 6 interfaces with a high of 20k pps when I looked a couple of minutes ago..Not untypical of the 30 bcn's and bln's on our network.. So the 4-6 Mb per peer thing is inaccurate. On the way high side.
Slot Total Used Free %Free ---- -------- -------- -------- ----- 6 61.67 M 32.82 M 28.84 M 46 %
I think the original figure was based on a peer sending a full table, which is pretty good IMHO. The figures above look a little on the high side, but i'd imagine ANS has quite comprehensive BGP policies in place from the RADB, and they would take up quite a chunk of memory. Besides, from recent experiences it looks like Bays bgp implementation doesnt return memory to the GAME engine once its unused, it just keeps it for future re-use. So the memory used actually reflects the most the BGP process has required since it started, not the current usage. Of course, the ANS code might behave differently. - jeremy -------------------------------------------------------------- Jeremy Hinton NOC - VisiNet jgh@visi.net "For years I was smart; I recommend pleasant." - Elwood P Dowd
From: Jeremy Hinton <jgh@visi.net>
AFAIK, ANS has stock hardware (FRE/ARE, with ARE performing soloist duties), same as us and other Bay cutsomers. Software is another issue though. Rob can probably shed some extra light on this, but i'll add what i've gathered.
Yep, this is all correct.
while mainstream progressed up to 11.x (w/ISP mode having *some* of the ANS addendums). 12.x is supposed to bring the code lines closer together, though i'd imagine ANS will still have needs that wont be rolled into the mainstream code.
Jeff Burgan's comments from yesterday summed it up pretty well, and he knows much more about this than I. I'd imagine your local bay guy would have the details of which stuff is in which release. RobS
participants (4)
-
burgan@corp.home.net
-
Chris A. Icide
-
Jeremy Hinton
-
Rob Skrobola