Master Dialogue: The host cache problem for ESXi is worth collecting!

Jar question:
We all know that ESXi 5 supports using SSD disk as host cache, and today I have a simple experiment, I use a Samsung 830 SSD disk, I found that can identify the drive type of SSD.
and the SSD host cache has been successfully enabled with this disk.
My understanding is that when the host memory is not enough, the need to use the VM swap files, the SSD host cache will be used first, when the SSD host cache space is not enough to use the virtual machine Exchange files.
Once set, the SSD cache is fully transparent when used, and does not affect the swap file settings of the VM or host.
Because my production environment is still 4.1, I can't actually verify the effect.
You are welcome to correct me.


Tigi replied:
When I first bought a third R710, I had intended to try the host cache, specifically configured for this SSD drive, but on the H700 card, VMware ESXi does not recognize that this is an SSD, ...
However, according to VMware's KB, you can forcibly specify that a hard disk is an SSD.


winxp2000 replied:
Can not use the cache as far as possible, the speed will not improve;
Of course, if your memory is always overflow, it is still equipped with a more secure
Also note that, by default, the virtual machine cache = You are equipped with the memory of this virtual machine
It is also placed in the file directory of the virtual machine by default and needs to be specified manually


Tigi replied:
Yes, I have thought about it, when I bought a new machine, 64GB of memory and 128GB of memory, the difference of less than 10,000, and buy an enterprise-class SSD, the price is not cheap-if for data reliability considerations, once needed to buy two to do the mirror ...
Or Cachecade is more cost-effective.


Jar Answer:
I test this machine is a desktop, SSD directly connected to the motherboard, no RAID card.
I estimate that if the SSD is connected to the RAID card, the host will not be able to obtain the SSD hardware information and can not be judged SSD.


Jar Supplementary Answer:
I have some esxi is not run key applications, is used to run test applications for many projects, the number of more, memory overflow can often reach 10G, this time can obviously see the CPU load and storage IO are relatively high.
If you specify the SSD as the storage for the swap file, it is actually wasteful and requires at least the sum of the specified virtual memory for all VMs. And because these VMs are decentralized, it's cumbersome to manage.
SSD Host cache can avoid the above problems, set up after the No tube, swap files are still built on the original storage, but as long as the memory overflow does not exceed the capacity of the SSD, such as 128G, the actual hard disk or swap files on the shared storage will never be used.
For example, my server is now the maximum overflow memory is not more than 10G, if you use a 64G SLC SSD host cache should be able to improve a lot


hokings reply winxp2000:
Why not use it as much as possible. There are two places in ESXi5 that involve caching issues:
1) host cache for ESXi (set in host cache Configuration)
2) Each virtual machine Exchange file cache (virtual machine settings-option-swapfile location settings), Taiwan's virtual Mr. Website suggested Swapfile put on SSD up, said will improve speed.
Put 1 and 2) at the same time on the SSD good.


Tigi replied:
If the memory is small, SSD generally-such as 32GB memory, 120GB SSD, because the host memory is not too large, so the total memory of the above is not too much, this time, the host cache and put the swap files on the SSD on the interesting/meaningful.
If the memory is large, the SSD is small-for example, 128GB of memory, 100GB SSD, how to use, then it is necessary to evaluate the overall (more advantageous as the host cache).
——————————————
Generally speaking, having SSDs is always good.


Jar Reply hokings:
I think if the host Exchange cache is enabled, there is not much need to put the swap file on the SSD again.
Take one of my servers here. For example, the physical memory is 96G, which has 81 VMS, total provisioned memory sum of almost 190G, that is to say, if these VMs are turned on, at least 190G of swap files are required.
Currently running a VM has 47, the total provisioned memory is 110G, at this time can see from the performance graph virtual growth of about 10G, swap operations occupy memory is probably 3.5G, that is, there is a 32G SSD host cache can absorb all the original need to use the swap file IO, Swap files are almost no use at all.
If the swap file is built on an SSD, the SSD must have at least 190G of space, and almost no IO, fully furnished.
For ESXi 4.1, which does not support host caching, it is the only option to build the swap file on an SSD.


Sifeng answer jar:
You mentioned 47 VMs, provisioned 110G of memory, virtual growth 10G, Exchange files 3.5G, my understanding is: physical memory + virtual growth + swap files = provision of memory, which is the virtual growth is memory overflow, this 10G size should be occupied SSD disk host cache now, that if the host cache is not configured, the 10G is the resources used, is the exchange of files.


Mengduo questions:
I set up the host cache, and then the swap file for the virtual machine does not change location, still on the original storage. Two questions to ask:
1. According to the above DX explanation, the virtual machine automatically uses the host cache to do the exchange, only when the memory overflows, will use the VM in the original storage Exchange file. No overflow will not be used, whether this understanding. However, by observing the file date in the virtual machine directory, we find that the date of the Exchange file in the virtual machine directory is the latest date, whether this can prove or use these exchange files.
2. Setting the host cache size should actually be roughly equal to the host physical memory size ... It's basically useless to be big. Consider that the virtual machine memory settings do not exceed the physical machine memory size too much.


Reply:
1. The date of the swap file is not updated to indicate whether it is being used, and VMware may automatically update some of the record information for the interchange file. Reliable judgment, it should be seen from the storage Performance Map to save the Exchange file storage above whether there is IO, if you specify a private LUN to save the swap file is easier to explain the problem.
2. The size of the host cache is not related to the size of the physical memory, preferably larger than the total provisioned memory of all VMS, so that the virtual memory of a basic VM can be stored in the host cache even if all swap out, it is guaranteed not to use the low IOPS swap file storage.


This dialogue is organized from http://bbs.vmsky.com/thread-39499-1-1.html


Another http://bbs.vmsky.com/thread-39869-1-1.html