Comments on: The Case Against FCoE and Fibrechannel – A Reasonably Complete View http://etherealmind.com/the-case-against-fcoe-fibrechannel/ Network design, architecture, thinking, working. Tech. Fri, 10 Feb 2012 18:43:00 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: JR http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-996 JR Wed, 10 Nov 2010 21:57:39 +0000 http://etherealmind.com/?p=1535#comment-996 Greg Thanks very much for posting this article on FCoE. I'm currently re-evaluating our storage strategy moving forward (we're currently a Fibre Channel environment) and find the idea of iSCSI and/or NFS an appealing technology for our VMware environment. We won't need dedicated fibre switches, can utilise our existing Cisco network switches and make the whole thing easier to manage. Your thoughts on why FCoE is not a good idea are very thought-provoking and it's good to see you expand on the occasional comment made on PacketPushers. JR Greg

Thanks very much for posting this article on FCoE. I’m currently re-evaluating our storage strategy moving forward (we’re currently a Fibre Channel environment) and find the idea of iSCSI and/or NFS an appealing technology for our VMware environment. We won’t need dedicated fibre switches, can utilise our existing Cisco network switches and make the whole thing easier to manage.

Your thoughts on why FCoE is not a good idea are very thought-provoking and it’s good to see you expand on the occasional comment made on PacketPushers.

JR

]]> By: Greg Ferro http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-995 Greg Ferro Sat, 25 Apr 2009 18:51:12 +0000 http://etherealmind.com/?p=1535#comment-995 I am thinking that Dell would like to offer FCoE in their mid range servers such as the1950/2950 series that are used by a lot of mid-size companies. People who may not perceive the value of the Cisco UF but want to move away from FC. I am thinking that Dell would like to offer FCoE in their mid range servers such as the1950/2950 series that are used by a lot of mid-size companies. People who may not perceive the value of the Cisco UF but want to move away from FC.

]]> By: Brad Hedlund http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-994 Brad Hedlund Sat, 25 Apr 2009 18:18:45 +0000 http://etherealmind.com/?p=1535#comment-994 If there is demand for a 1GE CNA as you predict, AFAIK, the hardware on Nexus 2148 is FCoE capable. It would just require the right knobs to be turned on in software. If there is demand for a 1GE CNA as you predict, AFAIK, the hardware on Nexus 2148 is FCoE capable. It would just require the right knobs to be turned on in software.

]]> By: Greg Ferro http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-993 Greg Ferro Sat, 25 Apr 2009 15:18:50 +0000 http://etherealmind.com/?p=1535#comment-993 While I agree with you, I think that many people will choose to implement FCoE on 1Gig because it is cheaper than purchasing a FC HBA. At the moment the vendors (Emulex/Qlogic) are focusing on the high value/high profit 10GB CNA market but it's only a matter of time until they release a 1GbE CNA. Although iSCSI would make more sense, some people will need to connect to their legacy FC arrays because they haven't upgraded them, or don't have support for iSCSI. The transition from block storage over FC, to block storage over Ethernet or IP isn't going to be easy. While I agree with you, I think that many people will choose to implement FCoE on 1Gig because it is cheaper than purchasing a FC HBA. At the moment the vendors (Emulex/Qlogic) are focusing on the high value/high profit 10GB CNA market but it’s only a matter of time until they release a 1GbE CNA. Although iSCSI would make more sense, some people will need to connect to their legacy FC arrays because they haven’t upgraded them, or don’t have support for iSCSI. The transition from block storage over FC, to block storage over Ethernet or IP isn’t going to be easy.

]]> By: Brad Hedlund http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-992 Brad Hedlund Sat, 25 Apr 2009 15:09:32 +0000 http://etherealmind.com/?p=1535#comment-992 Why would the Nexus 2148 need to support FCoE when there is no 1GE CNA available? FCoE doesn't make any sense at 1GE. Does FCoE makes sense with a 10GE fabric extender? You betcha :-) Why would the Nexus 2148 need to support FCoE when there is no 1GE CNA available? FCoE doesn’t make any sense at 1GE. Does FCoE makes sense with a 10GE fabric extender? You betcha :-)

]]> By: Jeremy Filliben http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-991 Jeremy Filliben Fri, 24 Apr 2009 13:25:43 +0000 http://etherealmind.com/?p=1535#comment-991 Greg, I'm glad to see someone presenting the other side of the story. FCoE certainly has its target niches, but iSCSI is clearly going to be the general purpose solution, when traditional NAS won't work. If in 5 years, FCoE is ubiquitous, we'll all have spent way too much money on our SANs! One small correction (sort of).. Nexus 2k does not currently support FCoE; it is only a 1gb Ethernet solution. I believe FCoE is on the roadmap for the 2k though. Thanks for writing this! Jeremy Greg,

I’m glad to see someone presenting the other side of the story. FCoE certainly has its target niches, but iSCSI is clearly going to be the general purpose solution, when traditional NAS won’t work. If in 5 years, FCoE is ubiquitous, we’ll all have spent way too much money on our SANs!

One small correction (sort of).. Nexus 2k does not currently support FCoE; it is only a 1gb Ethernet solution. I believe FCoE is on the roadmap for the 2k though.

Thanks for writing this!
Jeremy

]]> By: Dmitri http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-990 Dmitri Tue, 21 Apr 2009 21:36:30 +0000 http://etherealmind.com/?p=1535#comment-990 EMC Symmetrix, for one. Don't forget that FC is not only for Data Centre and is very actively used in high-end workstation environment - for example, a single uncompressed HD video stream takes up 1.5Gbps, which is already way over what 1GE can handle, and it is not unusual in video production environment to have more than one going at the same time. EMC Symmetrix, for one. Don’t forget that FC is not only for Data Centre and is very actively used in high-end workstation environment – for example, a single uncompressed HD video stream takes up 1.5Gbps, which is already way over what 1GE can handle, and it is not unusual in video production environment to have more than one going at the same time.

]]> By: mrz http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-989 mrz Tue, 21 Apr 2009 16:06:02 +0000 http://etherealmind.com/?p=1535#comment-989 What application and storage arrays drive more than 1GE? Mozilla has a 30 host ESX enviro and not a single host drives more than 100-200Mbps. What application and storage arrays drive more than 1GE? Mozilla has a 30 host ESX enviro and not a single host drives more than 100-200Mbps.

]]> By: mgk http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-988 mgk Tue, 21 Apr 2009 10:42:03 +0000 http://etherealmind.com/?p=1535#comment-988 that is exacty my experience... we are using the intel server GE NICs (with the upgraded firmware to enable iSCSI boot) and performance difference compared to iSCSI HBAs is very small, certainly not worth the addional money. (the intel desktop ones are not as good) If you use some cheap 5$ NIC without a TCP offload engine the difference will be a lot bigger though... that is exacty my experience… we are using the intel server GE NICs (with the upgraded firmware to enable iSCSI boot) and performance difference compared to iSCSI HBAs is very small, certainly not worth the addional money. (the intel desktop ones are not as good)

If you use some cheap 5$ NIC without a TCP offload engine the difference will be a lot bigger though…

]]> By: mrz http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-987 mrz Mon, 20 Apr 2009 23:54:01 +0000 http://etherealmind.com/?p=1535#comment-987 "If you havenít looked at hardware acceleration for iSCSI, then you should immediately do this." Surprised to read that - in our tests at Mozilla with modern 4-core servers, QLogic HBAs haven't made any difference in I/O performance. I've generally found them to be a waste of money with no noticeable gain. “If you havenít looked at hardware acceleration for iSCSI, then you should immediately do this.”

Surprised to read that – in our tests at Mozilla with modern 4-core servers, QLogic HBAs haven’t made any difference in I/O performance. I’ve generally found them to be a waste of money with no noticeable gain.

]]> By: Dmitri http://etherealmind.com/the-case-against-fcoe-fibrechannel/#comment-986 Dmitri Mon, 20 Apr 2009 23:16:58 +0000 http://etherealmind.com/?p=1535#comment-986 Greg, The cost of building adequately high-performance IP infrastructure to support massively scalable IP network (for iSCSI) will cost more than CEE/DCE. Your IP infrastructure will need something to ride on, and that "something" is, um, inevitably Ethernet, for your iSCSI you will need high-performance Ethernet infrastructure in place AS WELL as IP (which will ride on top of it). Really, the *only* reason for iSCSI is the low entry cost. As soon as you start hitting against your CPU limitations caused by HBAs without iSCSI hardware acceleration, you will run into exactly the same trouble as with FCoE - upgrage, upgrade, upgrade - HBAs, switches, routers, etc. I was frankly surprised to see you say that FC switches and ports are more expensive than Ethernet - firstly, FC had ports faster than 1G before 10GE came around and prices came down to anything resembling "reasonable", and secondly, price per port was very similar or cheaper than 1G Ethernet (for 2/4G FC). To summarize, in my opinion there is a place for both iSCSI and FCoE. First is for SMB, second is for those running out of juice on their 4/8G FC. Greg,

The cost of building adequately high-performance IP infrastructure to support massively scalable IP network (for iSCSI) will cost more than CEE/DCE. Your IP infrastructure will need something to ride on, and that “something” is, um, inevitably Ethernet, for your iSCSI you will need high-performance Ethernet infrastructure in place AS WELL as IP (which will ride on top of it).

Really, the *only* reason for iSCSI is the low entry cost. As soon as you start hitting against your CPU limitations caused by HBAs without iSCSI hardware acceleration, you will run into exactly the same trouble as with FCoE – upgrage, upgrade, upgrade – HBAs, switches, routers, etc.

I was frankly surprised to see you say that FC switches and ports are more expensive than Ethernet – firstly, FC had ports faster than 1G before 10GE came around and prices came down to anything resembling “reasonable”, and secondly, price per port was very similar or cheaper than 1G Ethernet (for 2/4G FC).

To summarize, in my opinion there is a place for both iSCSI and FCoE. First is for SMB, second is for those running out of juice on their 4/8G FC.

]]>