Ask the Experts: Virtualization & Cloud Computing Questions Answered

55 Comments

View All Comments

• marc1000 - Friday, March 18, 2011 - link

  I mean comparing servers that would have similar amounts of performance, but comparing fewer high-end CPUs versus a lot of low-end CPUs. like the 480 ARM-cores someone posted on that article i said.
  
  But I agree that for 16 core servers, using ARM/Atom would make sense if you have a light load. but only if you have sure that your load will stay low for a very long time. If you need to add hundreds of CPUs to keep up with the increase in usage, then it doesn't make sense.

• docbones - Thursday, March 17, 2011 - link

  Are there any updates, especially on the open source options, for support of 3d gaming in virtualized systems?
  
  I know vmware workstation has some support for when using the same local machine - but when using the server based product it doesnt seem like there is much out there yet.

• hoverman - Thursday, March 17, 2011 - link

  I spend a lot of time working with Citrix, and Citrix has some beta releases that support this using a couple of the Nvidia cards. It is coming, and citrix is THE leader in this type of technology. I figure by the end of this year we'll have something that is released from them.

• juhatus - Thursday, March 17, 2011 - link

  what if you are going to thousands of wm's, say one per user. you could be talking ten's of thousands. what are the tools to handle that amount?

• Guspaz - Thursday, March 17, 2011 - link

  The higher-end enterprise solutions from companies like VMWare can address this sort of scenario.
  
  VMWare Infrastructure, for example, can address up to 2000 guests running on up to 200 hosts. VMWare vCenter Server can address up to 3000 guests running on 300 hosts per instance. Beyond that, I'm not sure; VMWare's product lineup is insanely confusing. You can run multiple vCenter instances, but that's multiple clouds at that point.
  
  Eventually, it makes sense just to outsource to a cloud provider like Linode or Amazon rather than trying to do it yourself in-house. These companies don't necessarily provide a centralized management solution (beyond, for example, being able to manage all guests from a single web interface), but they'll take care of the infrastructure for you, and let you scale up to as many guest instances as your imagination desires. Netflix has done a number of presentations about why they decided to abandon their own infrastructure and move everything over to AWS that are quite insightful. Personally, I'd have picked Linode over AWS, but Amazon does have the advantage of having a bunch of hosted solutions (SimpleDB, S3) while Linode has concentrated on the core service, affordable, reliable, performant VM instances.

• bmullan - Thursday, March 17, 2011 - link

  re - VMWare Infrastructure, for example, can address up to 2000 guests running on up to 200 hosts. VMWare vCenter Server can address up to 3000 guests running on 300 hosts per instance.
  
  I don't think there is any one such number anyone can quote.
  
  We have to remember that those numbers are highly variable as it really depends on what kind of applications are being used, how cpu or disk i/o bound they are etc.
  
  example:
  If its just webservers then you get oneset of performance/capacity number.
  If its transaction Database servers then you'll get totally different performance/capacity number.

• James5mith - Thursday, March 17, 2011 - link

  As my company moves towards a virtualized network, we are looking to leverage infiniband, and its capability to transport multiple protocols to reduce clutter and increase throughput.
  
  I can find tons of information on initiators, HBA's, etc. I can find almost zero information on infiniband targets (SANS, NAS, etc.)
  
  The most useful information I've found so far is on zfsbuild.com and even that is limited. Any chance of a somewhat deep dive on infiniband and it's role in virtualization? Especially with the capability (and capacity if you use 40Gbps IB) to transport ALL protocols over a single (or redundant) link is very intriguing to me.

• Fleepo - Thursday, March 17, 2011 - link

  We're a relatively small credit union (150 employees). Because we're a credit union we don't use many cloud services (mostly due to percieved privacy concerns.)
  
  We run our banking solution inhouse on AIX hardware; most of our other IT services are now virtualised on Xenserver + a basic SAN for our live site, and direct-attached storage at our DR site.
  
  Contrary to a comment someone has made earlier, (eg Virtualisation without a SAN is unworkable) we've found it's been very useful on directly-attached storage to reduce the complexity and cost of our physical hardware, and setup and maintenance of our DR site has been simplified (eg, just copy the VM to the DR hardware and turn it on.)
  
  One thing that was really noticeable (and to me, annoying) was the dominance of VMWare in the market, and the reaction from many solutions providers when they learnt we *weren't* using VMware for our virtualisation solution. We've also found it difficult to find hardware and software certified to work with anything but VMWare,and a vendor willing to consider Xenserver or Hyper-V as a viable alternative to VMWare.
  
  So, my question is -
  Do you see the dominance of VMWare in the market as an issue, especially in terms of industry expertise?
  Do you consider virtualisation to be a realistic solution for smaller businesses? Again, some earlier commenters are suggesting you should just go cloud instead? (we're not doing this for various reasons, mostly political and regulatory)
  
  By the way, I like VMWare, however we can't justify the cost, given there are cheaper alternative solutions like hyper-V and Xenserver in the market. VMWare's aggressive marketing put me off, too.

• HMTK - Friday, March 18, 2011 - link

  Depends on what you call a SAN. Our customers are typically so small that going FC or 10Gb Ethernet would be a waste of resources. For those small customers we use HP MSA2000sa SAS SANs to which we connect up to 8 servers (or 4 dual path). No expensive switches necessary but still good shared storage.

• zephyrwind69 - Thursday, March 17, 2011 - link

  Here's some perennial ideas I review from time to time and run into all the time. Keep in mind we're in our 2nd generation of virtualization on VMWare and run 90% virtualized, blades, and a big vendor's storage system which is three letters and doesn't start with I.
  
  1) Simple Shared Storage Systems-Always comes up for DR and satellite offices. I'm talking about 2 SAS host arrays (e.g. Promise, etc.) or basic 2x host Fibre arrays. iSCSI @ 10Gb in a bake off for performance and features. You don't always need something from Data General.....Qualified vendor supported stuff.
  
  2) Perennial MS vs VMWare. You've touched on performance below but things have changed and I'd like to see VSphere 4.1 in a really good performance bake off.
  
  With CNAs in the picture:
  
  3) CNA @ 10Gig review (ties into #1)-Setup ease of use, configuration, and performance from Emulex and Qlogic under Dell, IBM, HP. Lotsa vendor-supplied reviews out there, few unbiased.
  
  4) Desktop Virtualization Round Up-Workstation class tools
  
  5) Virtual Desktops-How big can I scale and what's it cost. RemoteFX vs. Teradici. That kinda stuff....it's in it's infancy and a good ROI study would be good. Sure if you listen to vendors it saves millions over a lifecycle but what's the real hardware requirements?!
  
  BTW, I could care less about Xen but it's probably worth a token review :) When it catches up to the two big players I'll listen. I am a VMWare bigot though, it's good to have the rest and let everybody else chase you. You might need a Virtualization 101 based upon some of the feedback!