can be used on the mechanical hard disk?

[lizhichao]

If there is no SSD, yugabyte-db can be used on the mechanical hard disk?

[kmuthukk]

hi @lizhichao -

Unfortunately, no. YugabyteDB's design is not meant/optimized for hard disk environment. It requires SSDs.

[lizhichao]

Ok

[ddorian]

@lizhichao See: https://docs.yugabyte.com/latest/deploy/checklist/#disks
Optimizing for low latency and fast reads/inserts just can't work with HDD.
We have issues that will lower the overhead disk io:
Wal multiplex #3549
Tablet splitting #1004 (will make possible to start a table with just 1 tablet and grow from there)

After those land and by combining them with colocation + low number of tablets per-table + raid0 of some hdd, I will do some tests to see if HDD may be viable. But it will not be official though.

[tobiasmuehl]

@ddorian Looking forward to your tests.

The design of openIO (yes, I'm aware it's just a self-hosted S3 and not a full database) uses HDD for storage and requires a small SSD to be present for metadata that is accessed more frequently. Does Yugabyte have similar optimizations? Let's say we'd run each server with 4 HDDs and 1 SSD, would the SSD be used specifically for IO-intensive tasks or are all disks load-balanced equally?

[tobiasmuehl]

@ddorian From my understanding #1004 tablet splitting has landed. Do you think now is a "good enough" time to attempt using HDDs?

[ddorian]

Do you think now is a "good enough" time to attempt using HDDs?

No, would need wal multiplexing too. That's more important than tablet splitting.

The design of openIO (yes, I'm aware it's just a self-hosted S3 and not a full database) uses HDD for storage and requires a small SSD to be present for metadata that is accessed more frequently. Does Yugabyte have similar optimizations? Let's say we'd run each server with 4 HDDs and 1 SSD, would the SSD be used specifically for IO-intensive tasks or are all disks load-balanced equally?

It's a completely different workload. Think of it like this: you put the metadata of openIO INSIDE yugabytedb and put yugabytedb ON TOP OF ssd.

metadata in openio ==(they are actually the same thing!)== yugabytedb-data. Putting yugabytedb in HDD is exactly like putting openIO.metadata in HDD.

[ddorian]

As an example, seaweedfs is also an object store / s3, that is optimized for HDD. It's metadata, can reside in yugabytedb.