Will 4096 bytes alignment benefit on SSD with 512 bytes sector size?

1

My SSD physical sector size is 512 bytes. Will 4096 byte alignment benefit in IO operations? Currently the 1st partition (Recovery) has starting offset at 32 256 bytes according to msinfo32. The second partition (system) has the offset at 115 153 920 bytes.

Some more data from fsutil output:

Bytes Per Sector : 512

Bytes Per Physical Sector : 512

Bytes Per Cluster : 4096

Bytes Per FileRecord Segment : 1024

Spurlos

Posted 2016-03-28T23:38:09.873

Reputation: 11

What is the make and model of your SSD? – David Schwartz – 2016-03-29T00:55:11.013

Unfortunately, details of optimum alignment on SSDs vary from one make and model to another. In general, the same 2048-sector (1 MiB) alignment that's the default on HDDs works fine on SSDs, but there are exceptions to that rule. Sometimes a smaller alignment value might work as well, but the space saving by using, say, 1024-sector alignment is too trivial to risk the performance hit if you get it wrong. – Rod Smith – 2016-03-29T17:35:30.063

@DavidSchwartz PLEXTOR PX-256M5S (Marvell 88SS9174/9187) – Spurlos – 2016-03-30T00:22:25.653

@Spurlos Alignment can make a 40% performance difference on this chipset with small reads, especially when there are a lot of them at the same time. – David Schwartz – 2016-03-30T00:26:07.467

Simple answer: Yes, because internally the drive is almost certainly optimized for 4K alignment. SSD manufacturers know that modern Windows/MacOS/Linux align to to 4K (or multiple thereof) by default and, knowing this, they optimize for it. I don't have time to find references thus the comment only. You could always experiment and run benchmarks yourself. – misha256 – 2016-04-18T21:18:18.083

Answers

-2

No, all these numbers are fake mockups for SSDs. They are not organized in that structure, and it makes no difference, all bytes have the same access time, as no disk is rotating.

Aganju

Posted 2016-03-28T23:38:09.873

Reputation: 9 103

2This is very misleading. Two bytes that are in the same underlying physical sector can be read twice as quickly as two bytes that are in distinct physical sectors. SSDs are often IOPS limited rather than bandwidth limited and the number of IOPS required to perform an operation can significantly effect throughput. – David Schwartz – 2016-03-29T00:56:24.063

@DavidSchwartz I think that reasoning applies to alignment in general. The question was about 4K alignment specifically, which should not differ from 512B alignment in terms of IOPS? – jiggunjer – 2016-03-29T01:18:17.053

@jiggunjer It depends on the SSD. – David Schwartz – 2016-03-29T01:25:30.733

@DavidSchwartz That is too vague for me to understand. – jiggunjer – 2016-03-29T04:06:36.453

1@jiggunjer Different SSD chipsets map external requests to internal requests differently. The IOPS limit applies past the framing on some chipsets. – David Schwartz – 2016-03-29T16:12:52.623

@DavidSchwartz because some have 4k read blocks? – jiggunjer – 2016-03-30T01:13:18.087

@jiggunjer That's probably the most common reason. It's not really clear what's going on inside the chipsets, but the benchmarks don't lie. – David Schwartz – 2016-03-30T01:25:31.943