& cplSiteName &
Comments
Newest First  |  Oldest First  |  Threaded View        ADD A COMMENT
Page 1 / 2   >   >>
Pete Baldwin
Pete Baldwin
12/5/2012 | 5:20:46 PM
re: Arista's New Hardware Packs SDN


I honestly don't know and intend to ask around about it. 


I doubt there's a standard. My assumption would be that most single-box latency figures refer to port-to-port times under particularly favorable conditions (certain packet sizes, e.g.)


Anyone care to chime in with specifics about this, or even about how the marketing end is done?

patentchoi
patentchoi
12/5/2012 | 5:20:46 PM
re: Arista's New Hardware Packs SDN


Is there any standardized way to test the latency for these data center fabrics ? Or is there any 'Cisco math' at play here ? Where does this leave qfab whose latency is of the order of microsecs ?

rainbowarrior
rainbowarrior
12/5/2012 | 5:20:45 PM
re: Arista's New Hardware Packs SDN


LR,


Is it true that Arista's SDN, OF-enabled switches are several times more expensive than their non-SDN, non-OF switches?


And if so doesn't that defeat the whole purpose of SDN? Wasn't the main use case to create cheap, generic switchs?


 

Pete Baldwin
Pete Baldwin
12/5/2012 | 5:20:45 PM
re: Arista's New Hardware Packs SDN


The purpose of SDN isn't cheap switches, I would argue. That's a hoped-for consequence, but for carriers, it's not the raw purpose, and it probably wouldn't happen for a while anyway.


The purpose is to gain flexibility that the closed switch/router systems don't offer. Now, you could translate that into "We want generic switches," and that might be what the Stanford crew originally was after -- but the first generations of SDN will end up being implemented on name-brand hardware, in most cases.


If Arista's new switches are that much more expensive (the cheapest 7150 sells for $13K, according to the press release, so "several times" might be a bit much), it's more because of all the other things that are in hardware: NAT translation, IEEE 1588 support, and the VXLAN thing that required the use of Fulcrum.


That said -- OpenFlow can get expensive if you use a lot of TCAM memories, which is what the protocol was developed for.  So, people (IBM, e.g.) are designing OpenFlow-supporting switches that don't need so many TCAMs.

paradigm1974
paradigm1974
12/5/2012 | 5:20:44 PM
re: Arista's New Hardware Packs SDN


Disclaimer: I am a Juniper employee. I try not to be unduly biased but I'm only human.


I'm not sure if there is some sort of standardized and public methodology for measuring fabric latency but here's how Juniper does it.


QFabric's latency is measured from one edge of the fabric to the to other edge of the fabric. To say it another way, the latency metrics are the sum of the latencies of each of the parts of the fabric. The near-side QFabric Node, the QFabric Interconnect, the far-side QFabric Node and all 4 cables.


When you take a moment and consider everything that goes into the latency measurements we cite, I think you'll find that sub-3 to sub-5 microseconds isn't as long a time as it might have first seemed.


While our competitors make a lot of noise about how fast their individual switches are I think it's a better use of your time to focus on how long it will take for your data to go from one side of your fabric / data center to the other. Once someone considers that particular question, they'll see that QFabric is really, really fast.

RockyMtn_Jack
RockyMtn_Jack
12/5/2012 | 5:20:43 PM
re: Arista's New Hardware Packs SDN


Calculating Port-to-port latencies within a single switch are fairly well defined and understood, since this is a common benchmark. There is some level of marketing gamesmenship that comes into play with large parkets (if one counts the first byte of the packet, or the last) but generally port-to-port performance for a single switch is a fair measurment.


Once you get beyond the capacity of one switch (64 ports in this case) and calculate performace at a Data Center scale, performance is less repeatable and comparable between vendors. L2 & L3 performance can make a big differnence, and verdors can be counted on to pick a traffic mix that favors thier architecture, chassis switches cna have unpredicable latencies between blades and varying paths through the Access/Aggregation/Core layers means some flows makes only one hop, while others can make 5 or more.


In this light, Juniper's TOR switch compares with ~750ns Port-to-Port latency (which will presumably drop as the new chips from Broadcom & Fulcrum roll out) - and the QFabric (composed of the TORs + a central interconnct) latency of ~5 microseconds is probably the fastest/lowest way to connect more than 700 ports.

BigBro
BigBro
12/5/2012 | 5:20:42 PM
re: Arista's New Hardware Packs SDN


The Arista 7124SX (24 ports, 500ns latency, "non-SDN") has a list price of $12,995.


The Arista 7150S-24 (24 ports, 350ns latency, "SDN") has a list price of $12,995.


 

rainbowarrior
rainbowarrior
12/5/2012 | 5:20:41 PM
re: Arista's New Hardware Packs SDN


LR,


When a vendor claims to support "SDN" could you get some technical details as to what exactly they support? A lot of us are trying to separte the marketing hype and "SDN-washing" noise from real technology and innovation. "SDN" just seems to be popping up everywhere and seems at risk of losing any real meaning. 

rainbowarrior
rainbowarrior
12/5/2012 | 5:20:41 PM
re: Arista's New Hardware Packs SDN


Thank you for the clarification. I guess you have good information on the 7150S-24. So for Openflow 1.1 how many flows does it actully support and what is the flow setup rate? And when will it support OF 1.3 and what are the targets for it in terms of flow scale and setup/teardown rates?

BigBro
BigBro
12/5/2012 | 5:20:41 PM
re: Arista's New Hardware Packs SDN


> I guess you have good information on the 7150S-24.


All I know was in their press releases:


"Available immediately the Arista 7124SX switch starts at  $12,995 USD."


http://www.aristanetworks.com/...


 


"The Arista 7150 Series switches are orderable now and shipping in CQ4 2012, with a list price starting at $12,995."


http://www.aristanetworks.com/...

Page 1 / 2   >   >>


Featured Video
Upcoming Live Events
October 1-2, 2019, New Orleans, Louisiana
October 10, 2019, New York, New York
October 22, 2019, Los Angeles, CA
November 5, 2019, London, England
November 7, 2019, London, UK
November 14, 2019, Maritim Hotel, Berlin
December 3, 2019, New York, New York
December 3-5, 2019, Vienna, Austria
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events
Partner Perspectives - content from our sponsors
Edge Computing, the Next Great IT Revolution
By Rajesh Gadiyar, Vice President & CTO, Network & Custom Logic Group, Intel Corp
Innovations in Home Media Terminals for the Upcoming 5G Era
By Tang Wei, Vice President, ZTE Corporation
All Partner Perspectives