On Sun, Mar 6, 2016 at 3:16 PM, Heath Albritton <halbritt@xxxxxxxx> wrote: >> On Thu, Mar 3, 2016 at 10:17 PM, Christian Balzer <chibi@xxxxxxx> wrote: >> Fair enough. >> Sync tests would be nice, if nothing else to confirm that the Samsung DC >> level SSDs are suitable and how they compare in that respect to the Intels. > > I'll do some sync testing next week and maybe gather my other results > and put 'em on the web somewhere. > > >> Yeah, my curiosity was mostly if there is similar ratio at work here >> (might have made more sense for testing purposes to REDUCE the >> overprovisioning of the Intel) and where the point of diminishing returns >> is. > > I don't think you can reduce the over-provisioning below default > levels. I've played with the tools from Intel and Samsung and haven't > seen an option for this. In fact, the manufacturers aren't very > forthcoming about the fact that they are doing any over-provisioning. > > >> If I'm reading the (well hidden and only in the PDF) full specs of the >> 960GB 863 correctly it has an endurance of about 3 DWPD, so the comparable >> Intel model would be the 3610s. >> At least when it comes to endurance. >> Would be interesting to see those two in comparison. ^.^ > > I could test them side-by-side. Given the degree with which the SM863 > exceeds the performance of the S3710, I'm not sure the S3610 would > fare very well. Frankly, I think some of the limitations are in the > flash controller. I'm speculating here based on the spikes I've seen > on the Intel units that I've tested. How much difference is there regarding to performance between S3710 and S3610? > > As for endurance, the 845DC Pro was rated at 10DWPD. Again, I was > disappointed with the 3DWPD of the successor. Sorry, probably I don't understand this sentence properly. Do you mean that 10DWPD would not be good successor of 3DWPD which I'm thinking? Can you elaborate on this more? >I was reassured by the > folks at Samsung that with the same level of over-provisioning the > SM863 would have the same endurance as the 845DC Pro. > > Caveat emptor and all that, given that's not really documented > anywhere. This makes sense given how wear-leveling works. > >>> I think the stock kernel module is 4.x something or other and LSA, now >>> Avago has released P9 through P12 in the past year. When I first >>> started using them, I was on the P9 firmware and kernel module, which >>> I built from the sources they supply. At this point most of my infra >>> is on the P10 version. I've not tested the later versions. >>> >>> Everything is IT mode where possible. >> Yes, at least until kernel 4.1 the module was the 4.0 version. >> And I had no luck at all getting the newer versions into a generic kernel >> or Debian. >> And when I deployed the machines in question P8 was the latest FW from >> Supermicro. >> >> Kernel 4.4 does have the 9.x module, so I guess that's a way forward at >> least on the kernel side of things (which I think is the more likely >> culprit). > > I'm using Supermicro HBAs as well, but getting the drive directly from > Avago. On their site, I just look for the firmware for the 9300-8i. > I've successfully used their utilities to cross-flash a 3008-based HBA > from IR to IT mode. > > The package also includes an SRPM, which I've found to build > relatively easily. IIRC, some mods to the spec file were required, > but that's about it. > > > Cheers, > > Heath > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com -- Email: shinobu@xxxxxxxxx GitHub: shinobu-x Blog: Life with Distributed Computational System based on OpenSource _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |