Is a Synology brand NAS any good for hosting a vault?

[QStoss]

Thanks Gaulfinger - that is helpful information. Looks like you've found a great way to keep costs in check. I'd never thought of Nexsan with most folks advising EMC, Dell, NetApp and the likes of BlueArc. Have you heard of Nexenta?

[gaulfinger]

Nexenta? Yes and we have tested it on a limited basis in the lab. It looks like a terrific product, especially if you're into the deep dive performance trace capabilities or need the deduplication or replication features. (Asigra doesn't need those features since they are built into the product.) We found the per-TB license model on Nexenta to be a bit expensive. It starts at $200/TB per year and quickly doubles or triples when you add phone-in support like one would expect from a complete storage solution.

Nexenta is based on ZFS, which is available in open source from Solaris and FreeBSD, which can provide "free" versions if you're up to the task.

[mbottoms]

Q, keep in mind that storage performance is your most sensitive area with Asigra. Whatever you decide on, make sure that it can support big IO/ps numbers. Asigra will easily swamp lesser SAN and NAS offerings.

There is a tool that comes with the asigra software designed to evaluate storage.

[QStoss]

Mbottoms, good point. Gaulfinger... I have some interesting test figures here from a Data Robotics unit tested against a Nexenta running Asigra as the software on a 2 threaded test:
Data Robotics was creating approx 22 files/sec at a 89Kb/s rate
Nexenta was creating approx 371 files/sec at a 1.46Mb/s rate

[mbottoms]

I looked at Drobo for private cloud asigra solutions at one point. Their sales guy was very persistent. When I told him that his hardware just wasn't sufficiently capable, he ended up getting one of their engineers on a call with me. After about 5 minutes on the phone the engineer tells the sales guy that his hardware just isn't sufficiently capable. I was amused.

I do think that Drobo offers an excellent product at a good price. It just isn't designed for high intensity applications.

[gaulfinger]

I absolutely agree! Performance, especially as the system grows larger, is paramount. I don't think it's linear, either, More like logarithmic. To put this in perspective, our 43TB DS-System is averaging 45.1MBps of writes, according to iostat. Another system about 1/3 in size is cruising at 6.9MBps writes. Considering most of the writes happen at night (6pm-6am) you can expect that during peak times, writes are about double the average. These systems both have Nexsan storage.

QStoss, I would look for writes about 10-100x the numbers you're seeing if you want to scale to 10's of TB.

Also important is read performance. Empirically, Asigra requires 1.5 - 2x the reads for every write, especially if you heavily use the Autonomic Healing functions to validate data integrity. It also does concurrent reads and writes, which can be very hard on disk arrays. So many systems will favor writes and get very low read performance.

[Lasseoe]

Mbottoms, good point. Gaulfinger... I have some interesting test figures here from a Data Robotics unit tested against a Nexenta running Asigra as the software on a 2 threaded test:
Data Robotics was creating approx 22 files/sec at a 89Kb/s rate
Nexenta was creating approx 371 files/sec at a 1.46Mb/s rate

Nexenta is just software, which hardware was used and how was it setup ?

[gaulfinger]

Good point, Lasse. Was the Nexenta test fully equipped with ZIL and L2ARC on SSD? That makes a profound difference. RAID configs matter, too, especially with larger SATA drives. RAID10 maximizes write performance - makes a world of difference.

[Lasseoe]

I'd actually love to know the exact setup of that Nexenta system. I'm currently building a storage system with Solaris and ZFS, both read and write performance is fantastic, but as soon as I introduce NFS it plummets, a SLOG (ZIL on SSD) device does make a difference but it's still far from impressive. If I spread my I/O load over several zfs datasets and several nfs mount points it scales nicely, but seeing as DS-System, and most other software products, typically r/w to one mountpoint at a time it doesn't work so well. Of course, if I was to add say 10-15 ds-systems it would probably work out fine, but I am not

L2ARC shouldn't make a huge difference as most of the I/O is completely random, it might catch a few percent but that's it (I think), and seeing as data in L2ARC consumes memory I'd rather use that memory for my ARC.

[gaulfinger]

L2ARC does make a difference, as does physical RAM, to a ZFS system. There is a lot of meta data (directory structure, for example) that gets cached. Getting 10,000 IOps for that kind of data alone can bump up performance by 20%.

Our first production DS-System ran with Solaris/ZFS.