VMware Server Hosts - 2x8-core or 1x16-core?

VMware Server Hosts - 2x8-core or 1x16-core?

As someone who's been close to server design and VMware build-outs since the inception of this technology I feel I have something worthwhile to contribute on this topic.

Post Broadcom acquisition, the new VMware pricing is out. $50/core for 12mo on Standard (VVS) and $135/core for 12mo on Foundation (VVF). For most customers Standard or Foundation will be the way to go. For those enterprise shops that can utilize all the advanced features and have the funding to justify it, then vSphere Cloud Foundation (VCF) is your jam.

We've been using dual 8-core CPUs in VMware server hosts for quite some time - mainly for the following reasons:

  • Virtual server environments are different than VDI where you truly need the cores & GHz - for VSI (virtual server infrastructure) 8-core procs get the job done.
  • MS Windows Datacenter is licensed at a minimum of 16-cores per physical host, 2 x 8-core procs = 16 total cores.
  • Good blend of price/performance - 8-core procs typically have higher frequency

With the new VMware by Broadcom licensing model you are required to license 16-cores of VMware PER PROCESSOR. So if you have (2) 8-core CPUs in your VMware hosts you must license (32) total cores of VMware - this adds up quick for larger environments.

Many of our customers have been seeing 2-5X increases in renewals - mainly related to the 16-core requirement. Why VMWare didn't make this 8-cores per proc is beyond me - would have made a lot of sense to align with Microsoft, but things are what they are so we move forward.

All that being said, what impact would a single 16-core processor make in your environment v/s dual 8-core? Short version: half the recurring VMware costs

Here are a few things to consider:

  • Zero hit on vCPU to physical core ratio - you still have the the same amount of cores - we typically see environments that are less than 5:1 vCPU/physical core, honestly most are less than 3:1. Some are as high as 10:1 - workloads are different and mileage may vary here - IT admins typically know when they are CPU constrained - plenty of great tools out there to show this
  • Single CPU will get you access to half the server resources - i.e. server may have 32 DIMM slots but you can only use (16) of those - again, not a big deal for most - using 64GB DIMMs you can put 1TB RAM - want double that? then use 128GB DIMMs
  • Single CPU limits IO bandwidth - again, typically not an issue in modern servers that may have a few 10/25/100GbE uplinks and a perhaps one additional storage HBA
  • Upfront acquisition costs are less as well as recurring VMware licensing costs

The chart below illustrates a new build-out using (4) VMware hosts and 6400 series procs. List prices pulled from a mainstream server vendor. Easy to see $$ saved on hardware and long term significant savings on VMware. $17K (list) saved on hardware and $25K (list) saved over 3 years on VVF with single proc hosts.

One thing I can say with certainty is that customer IT budgets are not increasing at a rapid pace - honestly most are fighting for dollars and trying to figure out how to solve other problems (cybersecurity etc.). Customers should dig into their virtual server environments and plan host refreshes around what will best suit their needs - RVtools continues to be a free/simple way to get a solid finger in the wind on many measurements.

Final thoughts on VMware - one consistent question from customers has been "what else is out there?" - as a company, our virtualization base is 90%+ VMware. We don't see the maturity level out there with other products, not to mention an abundance of skillset around VMware, or the ecosystem built around the product. People aren't balking b/c the product is failing, on the contrary it's highly reliable and has an awesome track-record. If you can find ways to cut costs or change up your environment, in most cases the sticker shock quickly subsides - we've even had some customers that had a VMware cost DECREASE. Before jumping ship take a deep breath and consider the big picture - plenty of other pressing issues to pour cycles into.




Peter Daley

Enabling objective conversations with clients using their data

9 个月

@spencerhamons at NetApp has been running a free workshop program for several years helping clients optimize IT infrastructure. This Broadcom challenge is right up his alley. Contact him for an objective assessment with targeted recommendations at Spencer.Hamons @netapp.com

回复
Spencer Hamons, CHCIO, CFCHE, FACHE

Executive Architect & Strategic Advisor

9 个月

Great post here Jonathan, funny that we were having a similar conversation this morning. When doing our workshops over the past few years, everyone let VM optimization slide, even though we have been presenting that data for 5 years now. Suddenly, I'm getting calls from customers that went through workshops with us years ago asking for that data again. Of course we'll talk through it, but now it is a hot-button issue.

Jeff Merck

Director of Sales at CTG National

9 个月

Jonathan, great post! I have had this discussion with many customers. I think making the assessment on true TCO from 1) staying on VMware and right sizing compute relative to actual usage trends and truing up VMware licensing or 2) leaving VMware and going with another platform, you must not forget that you have to retrain staff and retool in many places to get the same outcome (no small undertaking). However, if you want to leave VMware based on not aligning with the new model (principle per se). There are ways to convert the files from vmdk to other hypervisor file types that do not require copying data if you are a NetApp customer. Glad to chat through this with any interested customers. However, I encourage you to do the actual TCO assessment for good measure. You may be surprised by the actual outcome of the data you uncover in the TCO process and what the true TCO is be that staying on VMware or moving to another hypervisor.

Some other ways to optimize further…. 1. Stop buying 1:1 core replacements. A modern Saphire rapids core with DDR5 is going to be orders of magnitude faster than that ancient skylake. 2. Faster storage. That hybrid system with 50ms of disk latency could be causing significantly higher cpu usage in aggregate. CPU threads waiting on disk I/O cause all kinds of second order effects. 3. Faster networking long. Flow control, and TCP retransmits are not kind to CPU usage at scale. RDMA for vSAN or VM traffic can shed more CPU usage. Stop using cheap onboard NICs. 4. Optimize SQL. VCF included all the vRealize content packs for database and query optimization. scream it with me at your DBA’s “Build an index and stop running SELECT * across the entire database!” 5. Balance and rightsize. Turn on DRS to max, and run vRealize right sizing reports. 6. Overcommit CPU and even RAM. If you are buying the best hypervisor scheduler on the market use it. Stop listening to hardware salesmen who say 1:1 vCPU to CPU ratios are best, while your hosts average under 20% usage. Test/Dev and idle VDI pools can overcommit ram. 7. Fix Scan Storms with hypervisor aware security. 8. Use storage snapshot offloads and backup offload APIs.

要查看或添加评论,请登录

Jonathan Peyton的更多文章

  • Tires – A Tragic Tale

    Tires – A Tragic Tale

    You are a tire manufacturer. Everyone loves your tires.

    14 条评论
  • Cybersecurity Training - #1 Weak Link

    Cybersecurity Training - #1 Weak Link

    In a recent LinkedIn poll, roughly (60) customers were asked which of the following (4) cyber-security elements were…

    11 条评论
  • 7/27/2020 - City of Lafayette - CYBER-ATTACK!

    7/27/2020 - City of Lafayette - CYBER-ATTACK!

    Got your attention? Before you start freaking out too bad, this is Lafayette, Colorado - not Louisiana. Does this make…

    1 条评论
  • The 5th Step

    The 5th Step

    With all of the ad-hoc summer high-school graduations, many of us are sending $$ and words of wisdom to young graduates…

社区洞察

其他会员也浏览了