|
Oct 20 |
How vCPUs Interact With Physical CPUs – Resources
Posted by Philip Elder on 20 October 2014 02:28 PM
|
|
Original Published Here: MPECS Inc. Blog: How vCPUs Interact With Physical CPUs – Resources Here are some excellent resources on how a hypervisor such as Hyper-V interacts with the CPU pipeline.
Essentially, having a bit of time while waiting for some things to complete I’ve done a bit of digging to figure out if the premise “All VM threads (vCPUs) must be processed in parallel” still applies to the CPU pipelines and architectures of today. Check out the conversation I’ve been having with Brian Elhert on his blog with the videos as it seems that the premise no longer holds true. There are other VM performance thoughts that we have had since day one that need to be tested or verified based on Brian’s responses. In our experience the following can have an impact on a VM’s performance:
Hat Tip: @BrianEh (Brian Elhert) Further reading on tuning Windows Server 2012 R2: Philip Elder Chef de partie in the SMBKitchen ASP Project Read more » | |
|
Jun 18 |
SOFS, Storage Spaces, and a Big Thanks to the Intel Technology Provider Program!
Posted by Philip Elder on 18 June 2014 05:18 PM
|
|
Original Posted Here: MPECS Inc. Blog: SOFS, Storage Spaces, and a Big Thanks to the Intel Technology Provider Program! What was once the Intel Channel Program and now ITP has been very generous to us over the years. We make no bones about our support of both the program but also the excellent Intel Server Systems and Intel Storage Systems that we deploy on a regular basis. With the introduction of the Grizzly Pass product line we received a product that was bang-on with Dell, HP, and IBM feature for feature, construction quality for construction quality, with two very significant advantages to the Intel product:
This is the product stack we have been working on lately for our Proof-of-Concept testing for Scale-Out File Server failover clusters, Hyper-V over SMB via 10GbE provided for by two NETGEAR XS712T 10GbE switches, and Storage Spaces performance testing.
The top two servers are Intel R1208JP4OC 1U single socket servers supporting the Intel Xeon Processor E5-2600 v1/v2 series CPUs. They have dual Intel X540T2 NICs via I/O Module and PCIe add-in card along with a pair of Intel RS25GB008 SAS HBAs to provide connectivity to the Intel JBODs at the bottom. Two of the Intel Server System R2208GZ4GC 2U dual socket servers were here for the last couple of months on loan from the Intel Technology Provider program. We have been using them extensively in our SOFS and Storage Spaces testing along with the other four servers that are our own. One of the Intel Storage System JBOD2224S2DP units in the above picture is a seed unit provided to us by ITP as we are planning on utilizing this unit for our Data Centre deployments. The other two were purchased through Canadian distribution. Currently two are in a dedicated use configuration with the third to be used to test enclosure resilience in a Storage Spaces 3-Way Mirror configuration. We have been acquiring HGST SAS SSDs in the form of first and second generation units with an aim to get into 12Gb SAS at some point down the road. We still have a few more first and second generation SSDs to go to reach our goal of 24 units total. The second JBOD has 24 Seagate Savvio 10K SAS spindles that will be worked on in our next round of testing. Our current HGST SAS SSD based IOPS testing average is about 375K on an 8 SSD disk set up in a Storage Spaces Simple configuration (similar to RAID 0):
We have designs on the board for providing enclosure resilient solutions that run into the millions of IOPS. As we move through our PoC testing we will continue to publish our results here. We are currently working with Iometer for our baseline and PoC testing. SQLIO will also be utilized once we get comfortable with performance behaviours in our storage setups to fine tune things for SQL deployments. Again, thanks to Scott P. and the Intel Technology Program for all of your assistance over the years. It is greatly appreciated. Philip Elder Read more » | |
|
Jun 9 |
Storage Configuration: Know Your Workloads for IOPS or Throughput
Posted by Philip Elder on 09 June 2014 11:30 AM
|
|
Original Here: MPECS Inc. Blog: Storage Configuration: Know Your Workloads for IOPS or Throughput Here we have a practical example of how devastating a poorly configured disk subsystem can be.
The above was one of the first Iometer test runs we did on our Storage Spaces setup. The above 45K IOPS was running on 17, yes seventeen, 100GB SSD400S.a HGST SAS SSDs. Obviously the configuration was just whacked. Imagine the surprise and disappointment one would have supplying a $100K SAN and ending up with the above results after the unit was put into production and the client was complaining that things were not happening anywhere near as fast as expected. What we are discovering is that tuning a storage subsystem is an art. There are so many factors that one needs to keep in mind as far as the types of workloads that will be running on the disk subsystem right through to the hardware driving it all. After running a large number of tests using Iometer, and with some significant input from fellow MVP Tim Barrett, we are beginning to gain some insight into how to configure things for the given workload. This is a snip taken of a Simple Storage Space utilizing _just two_ 100GB HGST SSD400S.a SAS SSDs (same disks as above):
Note how we are now running at 56K IOPS. Microsoft has an awesome, in-depth, document on setting things up for Storage Spaces performance here: We suggest firing the above article into OneNote for later reference as it will prove invaluable in figuring out the basics for configuring a Storage Spaces disk subsystem. It can actually provide a good frame of reference for storage performance in general. Our goal for our Proof-of-Concept testing that we are doing was around 1M IOPS. Given what we are seeing so far we will hopefully end up running at about 650K to 750K IOPS! That’s not too shabby for our “commodity hardware” setup. Philip Elder Chef de partie in the SMBKitchen ASP Project Read more » | |
|
May 26 |
Hyper-V 2 Node JBOD DAS Cluster IOPS Run
Posted by Philip Elder on 26 May 2014 07:07 PM
|
|
Original Posted Here: MPECS Inc. Blog: Hyper-V 2 Node JBOD DAS Cluster IOPS Run This is just the beginning of our testing with Iometer:
The setup is the following:
We had no idea as far as what to expect with this setup since each CSV is being managed by one node. As we go along learning how Iometer works and how the disks react to various workloads we will publish additional results. Then we will go on to run the same tests again only with the above setup configured in a Scale-Out File Server Cluster with a 10GbE backend facilitated by a pair of Intel Server Adapter X540T2 NICs in each node, NETGEAR 10GbE XS712T switches, and a pair of Hyper-V Nodes. Hopefully with the Active/Active setup we get with SOFS and SMB Multi-Channel our performance will come out a bit better! This test configuration was Option 1 in the Three Intel Server Systems based Hyper-V and Scale-Out File Server Clusters Post. Philip Elder Chef de partie in the SMBKitchen ASP Project Read more » | |
|
Dec 31 |
Hyper-V: Number of NUMA Nodes on a Dual Intel Xeon E5-2630 R2208GZ4GC
Posted by Philip Elder on 31 December 2013 07:39 PM
|
|
Here is a snip of the number of NUMA Nodes in a newly stood up Windows Server 2012 R2 Standard server with a pair of Intel Xeon E5-2630 CPUs in the Intel Server Systems Grizzly Pass 2U setup:
Here is the same setup showing the number of Cores/Threads:
Note that we do not have Hyper-Threading turned off on this particular server. It’s important to note that a VM that is set up with more vCPUs than cores on one CPU may actually perform poorer than the same VM set up with the vCPU setting equal to or one less than the number of cores available on one CPU. This is what is meant by spanning NUMA nodes. Suffice it to say we can spend a while discussing the performance impact of too many vCPUs assigned to one VM. Ultimately, one needs to stress test a VM setup using a variety of configurations to find what will be optimal for that particular VM. Happy New Year’s Eve everyone. All the best for 2014! Philip Elder Chef de partie in the SMBKitchen Read more » | |
|
Dec 23 |
Database Configuration for Maximum SharePoint Performance (Video, Slide and PowerShell Script)
Posted by Reprinted Article on 23 December 2013 10:41 AM
|
This post is way overdue. Since I’ve been getting a lot of requests about this specific presentation regarding SharePoint databases, …
Read more » | |
