- jeromel's home page
- Posts
- 2025
- 2020
- 2019
- 2018
- 2017
- 2016
- 2015
- December (1)
- November (1)
- October (2)
- September (1)
- July (2)
- June (1)
- March (3)
- February (1)
- January (1)
- 2014
- 2013
- 2012
- 2011
- 2010
- December (2)
- November (1)
- October (4)
- August (3)
- July (3)
- June (2)
- May (1)
- April (4)
- March (1)
- February (1)
- January (2)
- 2009
- December (3)
- October (1)
- September (1)
- July (1)
- June (1)
- April (1)
- March (4)
- February (6)
- January (1)
- 2008
- My blog
- Post new blog entry
- All blogs
NFSv4 performance
Updated on Mon, 2025-02-03 13:59. Originally created by jeromel on 2025-02-03 12:08.
Single node performance
Since we have a limited number of interactive nodes, this test tells how much things could slow down with concurrency on a single node.
To be sure the performance is indeed close to 600 MB/sec single thread, I did a dd test and it found 436 MB/s (I can believe in the 592 then).



Performance per thread drops rapidly with the number of threads. This means that users will feel a significant slow down as the number of users and operations will increase.
Single node performance
I used iozone for performance testing purposes but used a single node so far. Tests were simple, do 1, 2, 4, 8, 16 processes and measure average + graph.
Since we have a limited number of interactive nodes, this test tells how much things could slow down with concurrency on a single node. To be sure the performance is indeed close to 600 MB/sec single thread, I did a dd test and it found 436 MB/s (I can believe in the 592 then).
Processes | Readers | Reader / thread (MB/sec) | Initial writers | Initial writers / thread (MB/sec) |
1 | 188994.22 | 185 | 606677.81 | 592 |
2 | 507448.75 | 248 | 1827691.56 | 892 |
4 | 299481.93 | 73 | 2488218.81 | 607 |
8 | 357638.75 | 44 | 1407643.56 | 172 |
16 | 661923.67 | 40 | 1004270.04 | 61 |
Performance per thread drops rapidly with the number of threads. This means that users will feel a significant slow down as the number of users and operations will increase.
DD test from a few nodes
To make the result more accurate, 10 loops were made (11 GB file) and the tests were made from respectively 1, 2 and 4 distinct nodes. This shows a different picture. Performance seems fine a-priori and scales as a function of node (with error bars not presented in this cheap spreadsheet but the standard deviation is ~ 80 for each value).

»
- jeromel's blog
- Login or register to post comments