On Nov 18, 2015, at 12:16 PM, Chris Adams <linux at cmadams.net> wrote:> > Once upon a time, Warren Young <wyml at etr-usa.com> said: >> - They?re serious server-grade machines, not borderline flimsy boxes competing largely on price. Built in and supported from Silicon Valley, not China. :) > > iXsystems sells rebadged SuperMicro stuff, nothing special (not made in > Silicon Valley).Good to know, though I must say, the SuperMicro stuff I?ve used is a cut above typical desktop PC or commodity grade hardware. Not on par with super high end stuff, but well above average.> iX found and fixed a FreeBSD kernel NFS bug, but it was a > painful experience.I see that story in the exact opposite way: iXsystems found and fixed the problem, expending heroic levels of effort to do so. By contrast, I?ve had several $300-500 NASes become unmountable for one reason or another, and the vendor was no use *at all* in getting it remounted. I had to rebuild the NAS from backups each time. It?s rather annoying to buy a NAS, then later realize you need to buy *another* NAS as a mirror in case the first one roaches itself. Isn?t that what redundant storage is supposed to avoid? Meanwhile, I?ve never had a ZFS pool become unmountable, even when the disk enclosure hardware was failing underneath it.> Then, early this year, we had a node fail, and it took them almost a > month to get us a replacement.That?s not good. But have you gotten better turn time from the $300-500 NAS providers for the same service? Did you opt for advance replacement, and if not, why not?> Their idea of HA is to monitor the ethernet links, not the services;Do the $300-500 NAS boxes even try to do HA failover?> even though we have multiple links in a LAGI?ve also had trouble with FreeBSD?s lagg feature. Fortunately, my use case allowed me to switch to a round-robin DNS based load balancing scheme instead. I don?t think you can do that with NFS, by its nature.> And today, when trying to open a ticket, their website is broken because > one of their DNS servers is returning 10.0.0.240 for part of their > website (where the CSS is served).Yes, I noticed their site was running awfully slowly. Embarrassing, but I don?t see what it has to do with the quality of their FreeNAS boxes.
On 11/18/2015 11:55 AM, Warren Young wrote:> It?s rather annoying to buy a NAS, then later realize you need to buy*another* NAS as a mirror in case the first one roaches itself. Isn?t that what redundant storage is supposed to avoid?no, RAID is purely availability when faced with single or double drive failure, nothing else. classic raid is most certainly NOT about data integrity, as the raid stripes aren't checksummed, they assume hardware data integrity. -- john r pierce, recycling bits in santa cruz
Once upon a time, Warren Young <wyml at etr-usa.com> said:> I see that story in the exact opposite way: iXsystems found and fixed the problem, expending heroic levels of effort to do so. > > By contrast, I?ve had several $300-500 NASes become unmountable for one reason or another, and the vendor was no use *at all* in getting it remounted.So, I was offering my opinion (backed by some personal anecdotes) of iXsystems. The system we had with all this trouble was much more than $300 (more like $30,000); IMHO it isn't "heroic levels of effort" to do something they told us it could do before we wrote the check.> Did you opt for advance replacement, and if not, why not?Yes, we had purchased a support contract that included advance replacement. They had no replacement part and took several weeks to find one.> I?ve also had trouble with FreeBSD?s lagg feature. Fortunately, my use case allowed me to switch to a round-robin DNS based load balancing scheme instead. I don?t think you can do that with NFS, by its nature.Yes, NFS talks to a single IP at a time. My problem isn't with FreeBSD, it with the TrueNAS software; it considers any configured network link dropping as a reason to fail over (even if the link is in a LAG). That is not configurable behavior.> Yes, I noticed their site was running awfully slowly. Embarrassing, but I don?t see what it has to do with the quality of their FreeNAS boxes.Mainly just more anecdotal evidence about the company and their general reliability. I know there are fans of iXsystems and FreeNAS; I am not one of them (nor is anyone in my office). We also sold a TrueNAS system to a customer, they had trouble (different problems from us), and we just about lost the customer. -- Chris Adams <linux at cmadams.net>
On Nov 18, 2015, at 1:01 PM, John R Pierce <pierce at hogranch.com> wrote:> > On 11/18/2015 11:55 AM, Warren Young wrote: >> It?s rather annoying to buy a NAS, then later realize you need to buy*another* NAS as a mirror in case the first one roaches itself. Isn?t that what redundant storage is supposed to avoid? > > > no, RAID is purely availability when faced with single or double drive failure, nothing else. classic raid is most certainly NOT about data integrity, as the raid stripes aren't checksummed, they assume hardware data integrity.I knew I?d get some kind of lecture like that. Look, I know RAID/ZFS is not a backup. My point is simply that if you need to keep a mirror of your file server just in case it roaches itself, what you have there is dual redundancy, not a backup. You need an offline backup *on top* of that, for the same reason that all hot mirrors are not backups. My point is that unreliable NAS/RAID systems *require* this dual redundancy, whereas a reliable system only needs normal backups, that being the sort where you rarely go back and pull more than a few files at a time.