Benchmarks and Real-World are in two different worlds?

[el-John-o]

Hey all,

So a quick run down;

Machine is a small office free-standing IBM server I acquired from a family friend (a Doctor who owns said small office!). Figured it would make a decent NAS server, with the ability to hold more drives (later). It has 6 hot-swap bays and two CPU sockets (one CPU currently). It's old, but it's also free.

Running FreeNAS-8.3.0-RELEASE-p1-x64 (r12825)
Intel Xeon 2.8GHz Single Core
2GB of RAM
Currently;

2TB WD Green
1TB Hitachi 7200RPM

FreeNAS is installed on an 80GB HDD (the machine originally had 4 80GB drives in RAID5, it also only has USB 1.1, so this appeared to be the better solution. Not worried about losing the drive space. It's a dusty old SATA drive from '04)

I have them configured in a mirrored volume (yes, I'm aware I'll lose 1TB that way, but I'm not yet using 1TB of storage and the main purpose for this NAS is redundancy. When I get closer to 1TB I'll swap the 1TB Hitachi with another 2TB WD green)

Confession: Reading through the manual, reading through the forums, making an honest effort not to have to ask questions, but I am not yet finished with the manual!

Okay so here's the issue. Using hard disk benchmarks (like Black Magic Speed Test or Iometer) I'm hitting right around 60MB/s read and 50MB/s write speeds on my NAS. That's pretty good considering the age of the machine. I'm happy with the performance for now, in the future I'll build a faster machine perhaps.

But, in the real world? Not even close. It takes 30 seconds to 'populate' a folder (where I can see the contents on my screen) and transferring files is painfully slow. Previously, I was running Ubuntu Server, but I was attracted to the idea of FreeNAS because it seemed to do what I needed better, especially it's ability to manage volumes much easier than via Ubuntu Server. Benchmark performance went up, but real-world performance went way down

Any ideas what the issue is? Where to look? I'm using a CIFS share on both Windows and Mac clients. Windows doesn't natively read AFP, but Mac OS DOES natively access CIFS, so it just seemed easier to do it that way.

Presently, both data drives are connected directly to the motherboard, this server DOES have an IBM RAID controller. Would it be beneficial to connect the drives to the RAID controller? It's a PCI-X RAID controller, my thinking was it simplified things to leave the controller out of the equation since I was running the drives in JBOD mode anyway. If I switch the drives to the controller (just a matter of moving cables) will I lose any data?

Finally, and this is more of a secondary question; but I'm aware I'll never get lightning fast performance on this older system (I just want to get it at least as fast as it was under Ubuntu!). But I could potentially add a second Xeon CPU (I'm going to look too and see if this motherboard would support any multi-core Xeon chips). It has DDR2 RAM, that could be increased. It has no PCI-Express though (just PCI-X) so I'm limited on upgrading NIC's and the sort, as they are a bit harder to find and pricey. So, would there be any noticeable performance gains over adding some RAM and a second CPU? I'm satisfied with the performance, but I've seen the exact same 2.8GHz Xeon CPU go for $15 on eBay and RAM is cheap, so for a few bucks to get a bump in speed I'd go for it. Any more though, and I'd rather just put that towards building a new server altogether.

Thanks!

-John

 

[ProtoSD]

Take a look at this thread:

http://forums.freenas.org/showthread.php?5338-Browsing-directories-slow

 

[el-John-o]

Take a look at this thread:

http://forums.freenas.org/showthread.php?5338-Browsing-directories-slow

Thank you, I'm going through those tweaks listed now.

Do you have any comments on the other two issues though? It appears everyone in that thread had directory issues but still had good speeds. In mine, though it benchmarks with decent speed, file transfers are painfully slow.

Also, any ideas on whether upgrading this box would be worth it? Just curious.

Thanks so much for the tips!

-John

 

[ProtoSD]

Read through the thread carefully before you try the tweaks, there were some "adjustments" along the way. I think those tweaks will definitely help with the directory listings and help your transfer speeds a little. CIFS is a CPU hog, but it is also a single threaded (one cpu) app, so adding another CPU won't really help that. Adding more RAM should help though. You didn't mention if your mirror was ZFS or UFS.

Trying the tweaks is free, if they give you what you're looking for it's a pretty simple thing to try without spending money. If you're not satisfied after, I'd probably look at upgrading the system, though RAM never hurts.

 

[el-John-o]

Read through the thread carefully before you try the tweaks, there were some "adjustments" along the way. I think those tweaks will definitely help with the directory listings and help your transfer speeds a little. CIFS is a CPU hog, but it is also a single threaded (one cpu) app, so adding another CPU won't really help that. Adding more RAM should help though. You didn't mention if your mirror was ZFS or UFS.

Trying the tweaks is free, if they give you what you're looking for it's a pretty simple thing to try without spending money. If you're not satisfied after, I'd probably look at upgrading the system, though RAM never hurts.

It's ZFS.

Well, I've been messing with the tweaks for a while. Performance only got worse (I read every post in that thread). It finally got to the point where after 10 minutes I was staring at a blank window. So I reverted to a backed up copy of loader.conf and removed all of the auxillary parameters.

Bedtime now, I'll keep researching and looking through threads. I may need to just go back to Ubuntu server. I really like the idea of FreeNAS and ZFS but it doesn't seem to like me... Oh well, I won't give up just yet, I'll keep tweaking.

I've made a list of the most 'up to date' of all the tweaks in that thread, and tomorrow evening if I get a chance, I may start plugging them in one by one and seeing what works or what doesn't. I'm honestly okay with 60MB/s speeds and such, if the directory listings were just up to snuff and I could actually GET those speeds.

I still can't for the life of me figure out why a benchmarking app can pull 60MB/s, but I can barely breach 5 or 10MB/s in actual file transfers... (Windows or Mac)

EDIT: Well forget Samba! I did some more reading and though I'd try NFS. Vanilla, untweaked, unchanged, out of the box NFS and I'm pushing 115MB/s. So, I'm nearly saturating gigabit on a single core server from 2004 with some desktop class hard drives.

I guess I've found my solution!