Comments on: Five Myths About Blade Servers http://www.linux-mag.com/id/4153/ Open Source, Open Standards Sun, 21 Mar 2010 00:35:03 +0000 http://wordpress.org/?v=2.0.11 by: bongsf http://www.linux-mag.com/id/4153/#comment-2448 Wed, 22 Apr 2009 12:32:47 +0000 http://www.linux-mag.com/id/4153/#comment-2448 I never like the idea of having blades + VMware/virtualization for a fit all your 60 servers in 6U solution. Even though it sounded and looked cool, there're just to many drawback in it. Let's start with the hardware. As mentioned in the article, and contrary to what was written, IO is STILL a bottle neck. Just look at IBM's blade where there are limited to 2 Qlogic FC controllers per blade. This is already an indication it is never meant for high I/O demand. Remember we could squeeze 8 cores/32GB RAM in a blade, but we get a mere 8gbps total for disk access. This has not factor in the flexibility of design of SAN. As for the network interfaces where all blades are connected to a 6 ports Nortel switch. One really need to design the network carefully to segregate the bandwidth by configuring the switch for all the blades in order to not having all of them squeezing through a mere 1gbps uplink! As for virtualization, it's really about saving the money in hardware and send it to VMware. Look at the pricing of ESX, and it's RIDICULOUSLY expensive. Sometimes I just think there're too many trend followers dumping their money for a non-feasible solution. Where's the ease of management where all vm are running on its' own? And what is so much about hardware management cost? Do we expect to see a faulty component on all the individual server everyday? That's not including the case when one blade burn down 20 vm would just die at ONCE! When a 1u dies, ONE machine dies. I never like the idea of having blades + VMware/virtualization for a fit all your 60 servers in 6U solution. Even though it sounded and looked cool, there’re just to many drawback in it. Let’s start with the hardware. As mentioned in the article, and contrary to what was written, IO is STILL a bottle neck. Just look at IBM’s blade where there are limited to 2 Qlogic FC controllers per blade. This is already an indication it is never meant for high I/O demand. Remember we could squeeze 8 cores/32GB RAM in a blade, but we get a mere 8gbps total for disk access. This has not factor in the flexibility of design of SAN. As for the network interfaces where all blades are connected to a 6 ports Nortel switch. One really need to design the network carefully to segregate the bandwidth by configuring the switch for all the blades in order to not having all of them squeezing through a mere 1gbps uplink!
As for virtualization, it’s really about saving the money in hardware and send it to VMware. Look at the pricing of ESX, and it’s RIDICULOUSLY expensive. Sometimes I just think there’re too many trend followers dumping their money for a non-feasible solution. Where’s the ease of management where all vm are running on its’ own? And what is so much about hardware management cost? Do we expect to see a faulty component on all the individual server everyday? That’s not including the case when one blade burn down 20 vm would just die at ONCE! When a 1u dies, ONE machine dies.

]]> by: elittle http://www.linux-mag.com/id/4153/#comment-1762 Tue, 18 Nov 2008 16:44:33 +0000 http://www.linux-mag.com/id/4153/#comment-1762 I have worked with the HP Blade systems in conjunction with an HP EVA 4400. This has been a winning combination for me and my company. I can get more out of a blade system running ESX using boot from SAN than I ever could with a bunch of 1U servers ever could. I think that this is the way to go and the ROI is there. I have worked with the HP Blade systems in conjunction with an HP EVA 4400. This has been a winning combination for me and my company. I can get more out of a blade system running ESX using boot from SAN than I ever could with a bunch of 1U servers ever could. I think that this is the way to go and the ROI is there.

]]> by: srikanth.cheruku http://www.linux-mag.com/id/4153/#comment-1047 Tue, 06 May 2008 22:56:58 +0000 http://www.linux-mag.com/id/4153/#comment-1047 how much did you pay for the vm ware esx license? how much did you pay for the vm ware esx license?

]]> by: jskondel http://www.linux-mag.com/id/4153/#comment-1010 Fri, 18 Apr 2008 16:43:12 +0000 http://www.linux-mag.com/id/4153/#comment-1010 We currently have 21 Dell blades with RHEL 5 and WMVware ESX suite installed. Now we have over 110 VMware guests running and growing, ranging from Windows server to Linux to FreeBSD. We currently have 21 Dell blades with RHEL 5 and WMVware ESX suite installed. Now we have over 110 VMware guests running and growing, ranging from Windows server to Linux to FreeBSD.

]]> by: srikanth.cheruku http://www.linux-mag.com/id/4153/#comment-992 Tue, 15 Apr 2008 13:36:55 +0000 http://www.linux-mag.com/id/4153/#comment-992 I think, manufacturers know that blades are not that great, but they are just trying their luck by advertising too much. I think, manufacturers know that blades are not that great, but they are just trying their luck by advertising too much.

]]> by: gcreager http://www.linux-mag.com/id/4153/#comment-579 Fri, 09 Nov 2007 13:31:23 +0000 http://www.linux-mag.com/id/4153/#comment-579 Our experience with blades has been that while they're useful for HPC applications, they tend to concentrate heat discharge, making good thermal management a significant factor in planning and operations. Doing it over, I'd rather concentrate my systems in a multinode 1u chassis with more even rear-discharge air than a blade system with forced vertical exhaust. Our experience with blades has been that while they’re useful for HPC applications, they tend to concentrate heat discharge, making good thermal management a significant factor in planning and operations. Doing it over, I’d rather concentrate my systems in a multinode 1u chassis with more even rear-discharge air than a blade system with forced vertical exhaust.

]]> by: Jorge Reyes http://www.linux-mag.com/id/4153/#comment-571 Thu, 08 Nov 2007 19:40:54 +0000 http://www.linux-mag.com/id/4153/#comment-571 Blades Cost More On the cooling and power side Blades to cost more if you do know own your own data center. If are space on one of the largest data center providers they will charge you the same amount of money for power regardless if your servers can run on lower power when not fully used. On an outsource datacenter you pay full price for the maximum capacity of a 120/20 and the corresponding cooling that is associated with that power. That is: Amps * Volts = Watts Watts * hours = watthours watthours * 3.41214148 = Btu Blades Cost More

On the cooling and power side Blades to cost more if you do know own your own data center. If are space on one of the largest data center providers they will charge you the same amount of money for power regardless if your servers can run on lower power when not fully used. On an outsource datacenter you pay full price for the maximum capacity of a 120/20 and the corresponding cooling that is associated with that power. That is:

Amps * Volts = Watts
Watts * hours = watthours
watthours * 3.41214148 = Btu

]]>