This page provides some data around the Stash production instance that we run internally at Atlassian. We're providing this to give some idea of how Stash performs in a production environment. Please realise that this information is entirely specific to this particular instance – the details of your own installation may result in different performance data.
This data was collected with New Relic in February 2013, when the server was running a pre-release version of Stash 2.2.
On this page:
Hardware
The Atlassian Stash production server runs on:
- Virtualised hardware
- 4 Hyper-threaded cores
- 12 GB RAM
Load
Load data summary for February 2013:
| Type | Load |
|---|---|
| CPU usage | less than 30% on average |
| Load average | less than 3 on average |
| Physical Memory | peaked at 31% |
| Processes | Git: 17.3% CPU Java: 18.8% CPU |
| Clones | on average less than 300ms |
| Git operations/hour | peaking at 11,000 with an average of about 3,500 |
| Concurrent connections/hour | peaking at 100 connections with an average of about 40 concurrent connections |
| CI running against Stash server | 3 build servers with approximately 300 agents |
Server load
Git operations
Git clone operations
Git operations per hour
Git operations per hour (stacked)
Concurrent connections per hour
Git operations - cache hit/miss
Git operations - cache hit/miss
Git protocol usage per hour
Attachments:
Stash_server_2013Feb.png (image/png)
A_git-clone-duration-distribution.png (image/png)
B_git-ops-caching.png (image/png)
C_git-ops-lines.png (image/png)
D_git-ops-ref-advertisement.png (image/png)
E_git-ops.png (image/png)
F_git-protocol.png (image/png)
G_max-conn.png (image/png)