On Fri, Jan 19, 2018 at 11:25 AM, Marsell K <[email protected]> wrote:
> > My ZIL is never going to need more space than is required to store the > last ~5 seconds of write data, so even a 32 GB drive is going to go 90% > unused. > > Keep in mind that SmartOS is heavily used in datacenters for databases and > virtualization, which imply a high amount of synchronous writes. Take > virtualization: writes in KVM have to be treated pessimistically and thus > are synchronous and go to the ZIL. > > A ZIL doesn't take much space on a slog, but slog endurance is crucial > because if the device dies performance will fall off a cliff for many > workloads; performance is often an important part of correctness, so we > can't have that. Due to how SSDs work, all that extra space isn't being > wasted: it improves the device's endurance. > > Also, L2ARC is less useful than you'd think. Unless you've characterized > your workloads and/or done active benchmarking -- which is something only > you can do -- and found that L2ARC removes a significant bottleneck, > chances are it'll benefit your workload less than you hope. > > Yes, I'm aware that with TRIM support over time you are effectively using the entire "surface" of your SSD to improve your endurance. If that weren't the case, I wouldn't have done what I was doing, because I'd be reducing the lifetime of my ZIL devices by 90% or more. Only a few of my boxes use ZFS solely for virtualization; most of them also have a mixed workload of providing network storage to various clients via NFS or CIFS, and more recently using iSCSI as well. Obviously, L2ARC is going to improve the performance of those services. That said, I do have quite a few batch-type workloads that I could do reasonable benchmarking with, and I've taken the opportunity to do so. I rarely saw no improvement in total processing time for these jobs after adding L2ARC; in most cases, time to execute dropped by at least 15-20%, and some jobs improved by as much as 60-70% (only a few, to be fair, but I'll take the wins, regardless). Web serving throughput from my virtual apache systems almost always improved by at least 30%; I had simulated site-crawls that I did testing with, as well as some Selenium scripts I could use to repeatably test some complex form submissions and generation of web-based reports from MySQL backend databases. Finally, I have some video processing and transfer of the finished video files between storage locations, and I've found that this is always faster when I add L2ARC on the system where the processing is done and the result file will be *read* from for the transfer. So, yeah, for my workloads, L2ARC has always been a win. Sometimes a small win, often a *big* win, but *always* a win. best, Jim > Marsell > ------------------------------------------- smartos-discuss Archives: https://www.listbox.com/member/archive/184463/=now RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00 Modify Your Subscription: https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb Powered by Listbox: http://www.listbox.com
