![]() | |||||||||||||
Messages in this thread |
Hi, Jeff Garzik írta: > Alan Cox wrote: >>> That shouldn't be a problem, libata default DMA mask is 32 bits >>> (which isn't overridden with this controller) and so the block layer >>> will bounce any data being read/written above that point with IOMMU >>> or swiotlb. The comment is a bit unnecessarily scary. >> >> Adding a BUG_ON for this would be wise. Its trivial to check and a BUG >> rather than corruption if this assumption ever changes would be far >> preferable > > The default DMA mask -everywhere- is 32 bits. > > A lot of code will break if this assumption ever changes, not just > libata. > > Jeff thanks for clarifying this. I tested the effect of this patch on 2.6.22-rc2 + CFS-v13 with the current CVS version of PostgreSQL 8.3devel. pgbench with 25 clients and some large number of transactions to make the result stable showed substantial increase in throughput. Without NCQ, I got around 446 tps, with NCQ I got around 680 via local TCP connection. Previously, I got this level of performance only over local unix socket and smaller number of simultaneous clients. The disk is Seagate 320GB (ST3320620AS). Again, thanks for this patch. Best regards, Zoltán Böszörményi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ | ||||||||||||
| Last update: 2007-05-20 10:37 [from the cache] ©2003-2008 | |||||||||||||