Associate
- Joined
- 3 Apr 2011
- Posts
- 3
Hi all. I'll be getting a 120GB Vertex 3 (hopefully!) soon. I have a question regarding alignment of partitions to the physical erase block size, and whether this is still required for optimum performance.
The way I understood it, it used to be recommended to start partitions on a 512k boundary, so that a write of a logical OS block (4k) had no risk of spanning an erase boundary, and therefore requiring two erase/write's of the SSD to store it in the worst case scenario where the physical block required recycling. Presumably read performance in unaffected no matter what the alignment, as the drive is capable of reading any 512 byte sector it desires in linear time regardless of the physical block it lives in?
My question is whether this is still something to worry about, with the latest controller firmware (and with specific application to the sandforce chipsets). Presumably, LBA block addresses no long map in any way to the physical blocks on the raw silicon NAND that make up the SSD, and the firmware will redirect every write request (via on the fly compression) to the next available pre-erased block, and sequentially fill up each 512k block with whatever mix of LBA blocks is convenient, before doing an erase on the next one?
So if this is true, that seems to indicate that a partition not aligned to a 512k chunk (in the LBA view of the disk) has no bearing on what's happening on the silicon itself. Does this therefore mean previous advice to take care with partition alignment is now redundant?
The way I understood it, it used to be recommended to start partitions on a 512k boundary, so that a write of a logical OS block (4k) had no risk of spanning an erase boundary, and therefore requiring two erase/write's of the SSD to store it in the worst case scenario where the physical block required recycling. Presumably read performance in unaffected no matter what the alignment, as the drive is capable of reading any 512 byte sector it desires in linear time regardless of the physical block it lives in?
My question is whether this is still something to worry about, with the latest controller firmware (and with specific application to the sandforce chipsets). Presumably, LBA block addresses no long map in any way to the physical blocks on the raw silicon NAND that make up the SSD, and the firmware will redirect every write request (via on the fly compression) to the next available pre-erased block, and sequentially fill up each 512k block with whatever mix of LBA blocks is convenient, before doing an erase on the next one?
So if this is true, that seems to indicate that a partition not aligned to a 512k chunk (in the LBA view of the disk) has no bearing on what's happening on the silicon itself. Does this therefore mean previous advice to take care with partition alignment is now redundant?
