Hola, yo soy Elver Sena Sosa. Now
we're going to talk about disk groups, which is the unit of compute in a vSAN
cluster. These groups are actually a collection of disks that
exist in the hosts in that cluster. So, if you have an ESXi host, you will put a
bunch of disks inside the host. But, the disks themselves are put in
this collection, this logical unit of these groups, and in this group
itself there is an architecture where you have two levels. The first level
is the cache level where you have an SSD. There is just one disk, no more. And at this group there is only a single cache drive. The SSD is your cache. In the
second level, which is the capacity tier you can have either
SSDs or hard drives. Now, you can use either one of them, but for performance purposes you probably want to go with the SSDs. That will give you the best performance you can get. This second tier, or the
second level as you say, is called the capacity. In the capacity tier you
can have up to seven disks (a minimum of one and up to seven.) And, a single ESXi host can have up to five of these groups in a single host. Now if we were to do a quick map for vSAN, you can have seven of these on the disk group in the capacity tier and then one
cache. That means you can have eight different discs, being that this single
disk group times five. That is 40 disks that you can have in a single hose
providing storage to your vSAN and data store, 35 of which will be for the
capacity. Now you may ask yourself what's the role of the cache and what's the
role of the capacity disk? It all comes down to performance. VMware decided to design vSAN to be a very high-performance storage
solution. VMware does not have something like tier zero, tier one, tier two. vSAN does not have that. vSAN just has fast and super fast. How do you
differentiate between the fast and super fast? Well, when the write comes in to a disk group, it always goes into the cache.
Always, 100% of the time they write is going to come into the cache. The cache
is going to turn around and acknowledge that write. You get a write
performance of the cache speed 100% of the time. Now, the super fast difference comes in when the read comes in. When a read comes in, and if the block that you are reading is not in the cache, then the host that is hosting
to this group is going to go out to the capacity to find that block. Here is where the difference between a hybrid solution,
which is one where you have all hard drives, or an all flash solution makes
the difference. With a hybrid solution, you can see that when you have to read
from the capacity by being just regular spindles, your read is going to be a bit
slower than if you have an old flash, which would get better performance. Now I know the question you might ask yourself is, okay so the read comes into
the cache, it is not there but it's in the capacity. But, all the writes come in the
cache. So, when does the cache destage down to the capacity? Well, the cache is going to start copying blocks from the
cache into the capacity disk. When the blocks are cold, that's one way
that will start these destagings, which makes sense. And the other one is when the
cache starts running out of space. When the cache starts running out of
space or the capacity in the cache is about 90-plus percent utilization, the
cache would automatically start destaging stuff into the blocks
themselves, into the capacity tier of the cells, to make sure that we have enough
space here. And now one more thing I didn't mention is that something that vSAN does so the reads can at least be such fast as they can be, is that any time a read comes from the capacity tier for the hybrid, the cache will copy those
blocks and keep them in the cache so that the next time they read comes in for
that block, it will be in cache and the read will be responded directly from the cache. So, you get better performance that way. Something else about this is,
well we have hybrid we have all-flash, but what is different from the read? When the read comes in from the all-flash, the blocks that were read don't get copied back up to the cache. They stay in
the capacity. Because of that, there is a design decision that was made for the disk group on the cache. Whereas, when you have a hybrid solution
the cache by default is going to be split 70/30. So, 70% of the space in the
cache is reserved for the reading and 30% of the space in the cache is
reserved for the write. Now, on the all-flash, since there is no need to copy the blocks back to the cache, I mean this is SSD anyway so they are reading fast
regardless, there is no 70/30 split. It is 100% write only at the cache here. That is kind of the difference from a performance perspective between a hybrid solution and an all-flash. Now, a few things to add, if you go hybrid or all-flash every thing in this group in the cluster has to be the same configuration. It has
to be either hybrid or all-flash, you cannot mix and match those two. You can't do that. Also, the disk groups, the number of capacity disks that you have in any disk
group could be different, it could vary. However, it
is strongly recommended that these groups in the cluster are homogeneous. So, they have more or less the same number of capacity disks or
more or less the same cache disks, so that performance is consistent regardless of
where you're storing the information. One more thing is that when you have a
hybrid you will have your ESXi hose connect them to the network.
Well, you will connect it to the network regardless whether you have hybrid or
flash. But, when you have a hybrid one gigabit per second, it is supported for hybrid. When you have all- flash, it has to be 10 gig per second.
Of course if you go with the hybrid and you do 10 gig per second you will have
more capacity on your network, but still your slowness is going to be
when the capacity that reads come in. That's kind of it for what is a disk
group. I am Elver Sena Sosa, thank you for watching.
