Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update capacity planning doc #2295

Merged
merged 3 commits into from
Jan 18, 2024
Merged

Update capacity planning doc #2295

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e8383ef
:pencil: update capacity planning doc
vankichi Jan 18, 2024
9a7dac8
:pencil: Fix spelling
vankichi Jan 18, 2024
c100b2a
:pencil: Fix typo
vankichi Jan 18, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 17 additions & 6 deletions docs/user-guides/capacity-planning.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,34 +6,45 @@ Capacity planning is essential before deploying the Vald cluster to the cloud se
There are three viewpoints: Vald cluster view, Kubernetes view, and Component view.
Let's see each view.

<div class="notice">
When introducing production, we recommend that you actually measure how many resources are required for verification.
</div>

## Vald cluster view

The essential point at the Vald cluster view is the hardware specification, especially RAM.
The Vald cluster, especially Vald Agent components, requires much RAM capacity because the vector index is stored in memory.

It is easy to figure out the minimum required RAM capacity by the following formula.
The minimum required memory for each vector (bit) is:

```bash
// minimum required bits of vector
{ oid (64bit) + timestamp (64bit) + uuid (user defined) } * 2 + { dimension * 64 } + { the creation edge size + the search edge size } * 8
```

Considering the `index size` and `index_replica`, it is easy to figure out the minimum required RAM capacity by the following formula.

```bash
( { the dimension vector } × { bit number of vector } + { the bit of vectors ID string } ) × { the maximum number of the vector } × { the index replica }
{ minimum required bits of vector } * { the index size } * { index_replica }
```

For example, if you want to insert 1 million vectors with 900 dimensions and the object type is 32-bit with 32 byte (256 bit) ID, and the index replica is 3, the minimum required RAM capacity is:
For example, you want to insert 1 million vectors with 900 dimensions with 32 byte (256 bit) UUID, the index replica is 3, `creation edge size` is 20, and `search edge size` is 10, the minimum required RAM capacity is:

```bash
(900 × 32 + 256 ) × 1,000,000 × 3 = 8,7168,000,000 (bit) = 10.896 (GB)
{(64 + 64 + 256) × 2 + (900 × 64) + (20 + 10) × 8 } × 1,000,000 × 3 = 175,824,000,000 (bit) = 21.978 (GB)
```

It is just the minimum required RAM for indexing.
Considering the margin of RAM capacity, the minimum RAM capacity should be less than 60% of the actual RAM capacity.
Therefore, the actual minimum RAM capacity will be:

```bash
8,7168,000,000 (bit) / 0.6 = 145,280,000,000 (bit) = 18.16 (GB)
175,824,000,000 (bit) / 0.6 = 293,040,000,000 (bit) = 36.63 (GB)
```

<div class="warn">
In the production usage, memory usage may be not enough in the minimum required RAM.<BR>
E.g., there are a noisy problem, high memory usage for createIndex (indexing on memory), high traffic needs more memory, etc.
Because for example, there are a noisy problem, high memory usage for createIndex (indexing on memory), high traffic needs more memory, etc.
</div>

## Kubernetes cluster view
Expand Down
Loading