Talk:IOPS/Archives/2015

Latest comment: 8 years ago by Mattmill30 in topic IOps to MBs calculation

SSD examples

Table listing SSD examples should include the underlying base technology used.. (i.e. DRAM, NAND Flash, etc.) — Preceding unsigned comment added by 174.99.32.70 (talk) 16:53, 13 December 2011 (UTC)

  • Wouldn't it be more consistent to list types of SSD devices, (EPROM, SLC NAND, MLC NAND, DRAM, Memristor, etc) rather then a list of products? the hard disks are listed as classes (7200 RPM SATA) not specific brands/models (WD Raptor Black 7200) --74.81.156.51 (talk) 21:15, 23 December 2013 (UTC)
  • I agree, I feel like this table has stopped being a useful set of examples - the HDD table is a good rough guide to how IOPS translate to real world devices, but the SSD table is about 4 times too big and specific. I think the purpose of the table has been lost.--85.199.231.82 (talk) 11:59, 14 July 2015 (UTC)

IOps to MBs calculation

My understanding is that IOps is the number of blocks per second transacted. In which case, shouldn't the BytesPerSec calculation be (IOPS / BlockSize) * 8 = BytesPerSec? If not, please would you advise of the basis of IOPS or the calculation; and amend the Performance Characteristics section to provide clarity. Mattmill30 (talk) 10:42, 14 October 2015 (UTC)

Rules of Thumb

Currently there is no differentiation between 2.5" and 3.5" disks which have slightly different performance by shorter head paths and faster trsck reads. As a rule of thumb I heard a 2.5"/10krpm is similar to a 3.5"/15krpm concerning IO/s, while latency and throughput differ slightly (around +/- 10%). IO/s are "input/output operations" and they depend on size in bytes. It is easy to understand that smaller transfers need less time while big transfers (also named sequential) need more time. Minimum storage unit of rotating disks ist 512 bytes (larger ones, e.g. 8TB disks have 4096 bytes), while one transfer of SSDs is usually 4096 bytes. Transfer sizes smaller than this value show (nearly) no difference in the IO/s value. Since spinning disks read mostly complete tracks, there is little difference upto that value. Another rule of thumb tells us best "performance" is around 32kB IOs, where "performance" is defined by a trade-off between good IO/s and good througput. The third dimension of disk performance is latency. Concerning "disk systems" the forth performance dimension is the number of disks compared to capacity. The http://www.storageperformance.org uses an average io size of around 28kB (please check). You can see there different tests for streaming which means throughput (video) and iops & latency (databases). Some of this information shouldbe referenced in the eiki page.

User:aneuper 12:00 25 November 2015 (UTC)