Solved Zvol IOPS

I

ItMan

Hi,
I create a zvol from a pool with a disk 7200 rpm SAS 12g with avg latency 4.16 ms,
I present Vol to windows,
I run iSCSI initiator in windows and it add zvol to its disks,
I run IOmeter and select zvol disk , and in IOmeter only select, "default" workload, in specifications for workload, and run it ,
But it show for iops, it say 2000 iops and above,

I use zpool iostat -v 1
I show each secend that write to disk, without interrupt , and when I stop IOmeter ,it don't write any to disk,

Can you help me for iops , why it is very high?

Very thanks
 
ZFS does _lots_ of caching if it gets enough RAM and always tries to make writes sequential by accumulating small writes to bigger transactions. This can massively increase iops and R/W performance depending on the workload and pool configuration.

If you want to use ZFS as backing storage for VMs you should make yourself familiar with the underlying concepts and mechanics (e.g. caching and the intent log) of ZFS. ZFS can be very fast as VM storage, but with the wrong pool layout or configuration it can also become extremely frustrating to debug and work with. So I *highly* recommend you to at least read through the description section of the zpool(8) manpage to get a rough overview.
 
Hi again,
ok dear, my question is why with a disk 7200 rpm it show a high iops,
I want worry it use from cache, I need until get real IOPS,
while i set sync=always for zvol,

Thanks in advance,
 
why with a disk 7200 rpm it show a high iops
Click to expand...

I'm guessing English isn't your first language, but you're still not making much sense.

IOPS is I/O Operations Per Second; The number of reads or writes the disk can perform in one second.
2000 isn't a particularly high number (A single SSD can do 50,000 or more), and you want this number to be as high as possible.
 
I suspect he's wondering why it's 2000 when a typical 7200 RPM SAS disk has, on average, around 150-200 IOPS. But to be honest, it's really not that interesting to know what a single disk does. It's much more valuable to know what the IOPS of the entire chain is (network, controller, disks and all).
 
SirDice said:
why it's 2000 when a typical 7200 RPM SAS disk has, on average, around 150-200 IOPS.
Click to expand...
yes, it is exactly my question,
1. I read about sync writes, zfs return ack to application when data write to disk,
but in action , it is not work , because result for IOPS is higher of a disk,

2. do ZFS use RAM for cache sync writes,

sko say:
sko said:
ZFS does _lots_ of caching if it gets enough RAM and always tries to make writes sequential by accumulating small writes to bigger transactions. This can massively increase iops and R/W performance depending on the workload and pool configuration.
Click to expand...

3. it is mean, that first zfs hold sync writes to RAM and write to disk,
so, what is different it with async writes?

I'm sorry for long my question,
very thanks,
 
I have no idea what kind of workload IOmeter creates. Why does this matter? There are some workloads for which normal spinning disks (like the 7200 RPM you are using) can achieve way more than 100-200 IOps. For example sequential reads of small blocks.


Why are you worried about caching? Please explain why you insist on seeing the disk's raw IOps value, when in reality you are measuring a whole complex stack.

To make matters more confusing: There now exist spinning disks that have higher IOps, by using a write-cache inside the disk. Don't worry, that write cache is persistent, even on power failure.
 
Most Spinnig Rust and SSD manufactures have drooped their 7200KRPM product lines and opted for packing denser data volumes into their 5,400 RPM drives forall the Server,Desktop and Laptop Drive.
 
That might be true in the consumer space (in particular 2.5" drives), where I have little experience. But the bulk of the world's disk drives go into the data center, and there 7200 RPM seems to still be by far the most popular speed. And since IOps is getting a relatively larger problem compared to capacity (as capacities increase and are less of a bottleneck), I don't expect this to change very soon.
 
Chipper8827 said:
Most Spinnig Rust and SSD manufactures have drooped their 7200KRPM product lines and opted for packing denser data volumes into their 5,400 RPM drives forall the Server,Desktop and Laptop Drive.
Click to expand...

Both Seagate and Western Digital have large portfolios of 7200 RPM 3.5" drives, covering desktop, NAS, nearline, bulk storage, and archive roles. And their subsidiaries Hitachi, and Toshiba also have 7200 RPM drives. Ranging from 0.5 TB to 14 TB, using normal PMR platters, annoying SMR players, and even new methods for laying down magnetic tracks on platters. Using both SATA and SAS interfaces.

7200 RPM drives are nowhere near dead yet. Not until you can buy 14+ TB SSDs for under $0.50/GB, which isn't going to happen anytime soon. Lol.
 
SirDice said:
Ehm, really? There are 5400 RPM SSDs?
Click to expand...

I understand some trolls are fun, but you really ought to ban that one: posts contain just enough technical gibberish to look believable on first read, yet they never make any sense. There is plenty confusion with normal users as it is, we don't need intentionally misleading information.
 
You must log in or register to reply here.
Back
Top