How many of you use small servers ?

[taz722]

I think the new licensing policy negatively impacts the customers with smaller servers. However, is it not true that as part of consolidation, customers are moving to more dense servers with more CPUs per chassis. If  I have 8 sockets on my servers and with the limit of 96 GB per CPU , my total entitlement for the server is 96 * 8 = 768 GB, which seems prett high. I think where this licensing hurts is when customers have single/dual socket servers. But,aren't we moving away from those to more powerful servers. What proportion of your servers are small vs. large ?

Thanks.

[mcowger]

Many of my customers dont want to move to servers that big because they dont want that many eggs in a basket.  If you are running, say, 50 VMs on a big box like that, when (not if) it dies, you lose 50 VMs.  Most of my customers would prefer to use 2 4 socket systems or 4 2 socket systems and have fewer eggs/basket.

[mittim12]

We prefer the scale out approach to the scale up for the reasons that Matt stated.  We always felt it would be better to drop 30 or 40 machines instead of 60 or 70.   We are currently using 2 socket quad core cpu. 

On our desktop infrastructure we took a different approach and decided to drive for the highest consolidation ratio we could get.  

Message was edited by: mittim12 to correct spelling.

[mcowger]

Good point - the people that I *do* see using the ginormous machines are doing VDI with them.

[taz722]

Even with small servers , say dual socket, don't you think the current vRAM entitlement is more than enough. If you have a dual socket server, say 8 cores each. You vRAM entitlement for the entire server with E+ is 96 * 2 = 192  GB. If you create the maximum number of VMs you can on the 16 cores, that is about 12 GB per single core VM. Does anyone think that that's not enough ??? I honestly think that the original entitlement was slightly low, but with 96 GB for E+ , I don't see any reason to complain.

Of course not everyone uses E+, but then if you are using E or standard, you are likely to have smaller workloads that probably are not as memory intensive.

Comments anyone ??

[GaryHertz]

No, I don't think 96GB per socket 12/GB core is enough.  My current servers are using 128GB/socket, 8 core/socket or 16GB/core.  I'm not using all of that now but I plan to at some point in the future and 4.1 will allow it.  I'm currently using 15% CPU and 30% memory so I will run out of memory before CPU.

My previous servers were at 96GB/socket but were only 4 core sockets as opposed to 8 core.  I ran out of memory on these servers before I ran out of cpu.  96/4 = 24GB/core.  If you multiply that by 8, my current servers should have 192 GB/socket to match the potential of my old servers.

[taz722]

Wow, then you must be one of the few VMware customers who are 90 - 100% virtualized. Is that so ? What apps are you virtualizing that you have such high mem needs. I am sure that's probably not the case for majority of your apps. Correct ?

What do others have to say to this ? What are your average mem / Virtual Instance requirements ?

Thanks.

[wdroush1]

taz722 wrote:

Wow, then you must be one of the few VMware customers who are 90 - 100% virtualized.

Wait, this is abnormal? I thought only paranoid people kept more than a handful of boxes physical.

[kcucadmin]

I can honestly say we are 100% virtual.  i have not found anything that can not run as good in vmware if not better, if allocated the correct Disk I/O, and Resource Reservations.  I'm not saying that EVERYTHING can be virtualized.  i'm just saying as a fairly large Urology Practice with over 70 VM Servers, we have had "ZERO" performance issues related to Virtualization.  ( Well if you ignore the insanity that was trying to run all that on NFS data stores, once i switched to iSCSI all was good).

currently I have 8 hosts, 2 socket 4 cores w/HT (16 logical proc per host) with 96GB ram ea, allot of my hosts are running at that 70% memory mark, i was planning on doubling RAM in the next year or two once 8GB dims became a little more reasonable.  our CPU utilization is nill. 

Honestly i'll have 1gig iSCSI (currently 4 nics dedicated to iscsi per host) congestion long before i'll have CPU congestion.  i was looking at upgrading to 10gige.

frankly i underestimated how dense we could go per host.  i am looking at shutting down 2 hosts now and not refreshing that hardware down the road.

[GaryHertz]

taz722 wrote:

Wow, then you must be one of the few VMware customers who are 90 - 100% virtualized. Is that so ? What apps are you virtualizing that you have such high mem needs. I am sure that's probably not the case for majority of your apps. Correct ?

What do others have to say to this ? What are your average mem / Virtual Instance requirements ?

Thanks.

I'm about 70% virtualized.  Most of the people I've talked to are at the level or above.  From my experience the majority of people are currently at or plan to get to the 90-100% mark in the near future.

Most of my servers have 4GB of memory. I have 5 or 6 that have 6GB-8GB.  I think I have three SQL servers that use 16GB.  I would like to virtualize my remaining servers that have 16 to 32 GB each in them.  I've been to more than one VMware sponsored seminar that focused on the benifits of converting these high memory SQL and Exchange servers to VMware.  To convert them will now cost me about $10,000 in licensing and SNS under v5.  That's pretty hard to justify when they will cost nothing to leave them on the current servers or under v4.  I won't be able to stay at 4.1 forever because eventually it will become obsolete.  Any long term planning has to take into account the new license model.

I know every environment is different. My initial thought was why do you need such high amounts of cpu.  I think my environment is pretty typical of a Windows environment.

[wdroush1]

Gary H wrote:

taz722 wrote:

Wow, then you must be one of the few VMware customers who are 90 - 100% virtualized. Is that so ? What apps are you virtualizing that you have such high mem needs. I am sure that's probably not the case for majority of your apps. Correct ?

What do others have to say to this ? What are your average mem / Virtual Instance requirements ?

Thanks.

I'm about 70% virtualized.  Most of the people I've talked to are at the level or above.  From my experience the majority of people are currently at or plan to get to the 90-100% mark in the near future.

Most of my servers have 4GB of memory. I have 5 or 6 that have 6GB-8GB.  I think I have three SQL servers that use 16GB.  I would like to virtualize my remaining servers that have 16 to 32 GB each in them.  I've been to more than one VMware sponsored seminar that focused on the benifits of converting these high memory SQL and Exchange servers to VMware.  To convert them will now cost me about $10,000 in licensing and SNS under v5.  That's pretty hard to justify when they will cost nothing to leave them on the current servers or under v4.  I won't be able to stay at 4.1 forever because eventually it will become obsolete.  Any long term planning has to take into account the new license model.

I know every environment is different. My initial thought was why do you need such high amounts of cpu.  I think my environment is pretty typical of a Windows environment.

The only way our virtalization plans are even somewhat financially reasonable is we can buy essentials, otherwise we would be looking at the same issue where we point at a VM and go "that will cost $x,xxx to virtualize that one server!".

[wdroush1]

The problem isn't so much >2 sockets, it's how aged your hardware is, right now I can buy an Opteron board and cram a 1 socket CPU and drop up to 256GB of RAM on it, in a few months AMD is coming out with even more dense Opterons (16-core). That is a one socket server that goes beyond the licensing scheme many times over.

Sure, when you're looking at motherboards of yesterday, it's fine, when you're in the market for refreshing hardware or your first buy, it's terrible.