What Performance to expect from a C2558 (A1SRM-2558F)

[niaaard?]

Hello,

I have C2558 Motherboard at my disposal. I crammed some HDDs in it just to look how the setup would perform with FreeNAS.

What Performance can I expect from a CPU like that?

The Setup was:

A1SRM-2558F
16GB RAM
1 to 4 HDDs (Enterprise Grade 7200rmp 2,5")

No matter how I configured my vdevs (one mirror, two mirrors in one pool, RAIDZ1, RAIDZ2), I always got only ~40MB/s read AND write speed over CIFS.
What I also did was iperf3 test. Sending to the NAS would give me only ~600MBit/s, receiving from it would give the expected 950Mbit/s wirespeed. The former also would make the CPU spin a bit, while the latter showed no load on the CPU (of the NAS).
Also it didn't make a difference if the pool was encrypted or not. I guess thats due to the AES-NI of the board.
I didn't make more tests until now.

I realize, that this is no powerful machine, but I really expected more from it. I just want to do a sanity check here in the forums: Am I expecting to much?

As it is possible for me to use the machine as FreeNAS Server with up to eight HDDs, I want to make sure that it will be a useful setup. I really would like to have wirespeed writespeeds on CIFS (syncronous writes ON).

 

[hugovsky]

CIFS is single-threaded so, low power cpus can be a bottleneck. I have A1SAi version. I can saturate Gb link with CIFS. I think the problem is more with your network(maybe). Can you give more details of network and other system used to test?

 

[niaaard?]

Thx for your fast reply!

HDDs at my disposal:
Seagate 2.5" Enterprice Capacity 2TB
WD 2.5" Velociraptor 1TB

TP-Link 8/10-Port managed Switch TL-SG3210 with no load on it

I have several clients I could use, but I used one of my more powerful workstations:
Supermicro X10-SAT
Xeon E3-1241 v3
32GB RAM
LinuxMint 17.2 x64
not to forget: Samsung 840pro 512GB
I used the onboard NIC Intel i217LM (I also could use the included i210at, but that shouldn't make a difference)


All involved boards have their newest Firmware flashed.

I didn't do FTP or NFS benchmarks, as don't want to use them "in production", but ofc i could try them for troubleshooting purposes.

 

[hugovsky]

Well... your network and test system specs seem ok. Next step is to benchmark disks and pool.

check here:

https://forums.freenas.org/index.php?threads/performance-testing-with-dd.351/

https://forums.freenas.org/index.php?threads/notes-on-performance-benchmarks-and-cache.981/

 

[Ericloewe]

Do note that Samba is singlethreaded per connection. If you're talking to 3 clients, each will have their own thread (at the very least).

 

[niaaard?]

I am aware of that. Most of the time there is only one connection with high bandwidth demand.

But I'm doing more investigation right now. Benchmarks and other stuff...

 

[DJ9]

Here's a related thread regarding the same motherboard.
https://forums.freenas.org/index.php?threads/atom-c2558-a1srm-2558f-freenas-compatibility.17462/

 

[niaaard?]

According to that thread there shouldn't be a CPU bottleneck.

dd Benchmarks on a Pool out of 2 1TB Velociraptor mirrors (4HDDs):

Code:

dd if=/dev/zero of=tmp.dat bs=2048k count=50k
51200+0 records in
51200+0 records out
107374182400 bytes transferred in 325.127047 secs (330252999 bytes/sec)

So thats 315MB/s writing... sane?

Code:

Code:
dd if=/dev/zero of=tmp.dat bs=2048k count=50k
51200+0 records in
51200+0 records out
107374182400 bytes transferred in 325.127047 secs (330252999 bytes/sec)

Which results into 438MB/s.

 

[cyberjock]

The c2xx8 series lacks a lot of CPU instructions like SSE and others that are frequently used. I wouldn't be too surprised if performance overall just isn't what its cracked up to be. CPUs like the c2558 allow for a new Intel tech (forget the name) where purpose-built systems with purpose-built code that uses the new tech can see some amazing performance gains. But you won't see those gains with FreeNAS since it's a niche market for that kind of workload.

I'd recommend you go with the standard 2550 CPU instead of the 2558.

Of course, I don't know if anyone has provided any insight into how the 2558 performs in the past, so I can't tell you if my guess is right or not.

 

[niaaard?]

I'd recommend you go with the standard 2550 CPU instead of the 2558.

That box wasn't planned for that use in the first place. I just was curious if I can use it for FreeNAS. Obviously I can, it may must be a bit too slow. So maybe it will be used as backup server.
If I wanted it to be my fully fledged SOHO-NAS I'd rather stick to your usual recommendations... (Xeon blablabla) :)

 

[Jailer]

Or sell it and get something geared more towards what you are trying to do. That would make the basis of a nice fast pfsense box for someone.

 

[niaaard?]

dmesg from FreeBSD 10.1

Code:

CPU: Intel(R) Atom(TM) CPU  C2558  @ 2.40GHz (2400.11-MHz K8-class CPU)
  Origin = "GenuineIntel"  Id = 0x406d8  Family = 0x6  Model = 0x4d  Stepping = 8
  Features=0xbfebfbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,CLFLUSH,DTS,ACPI,MMX,FXSR,SSE,SSE2,SS,HTT,TM,PBE>
  Features2=0x43d8e3bf<SSE3,PCLMULQDQ,DTES64,MON,DS_CPL,VMX,EST,TM2,SSSE3,CX16,xTPR,PDCM,SSE4.1,SSE4.2,MOVBE,POPCNT,TSCDLT,AESNI,RDRAND>
  AMD Features=0x28100800<SYSCALL,NX,RDTSCP,LM>
  AMD Features2=0x101<LAHF,Prefetch>
  Structured Extended Features=0x2282<TSCADJ,SMEP,ERMS>
  VT-x: PAT,HLT,MTF,PAUSE,EPT,UG,VPID
  TSC: P-state invariant, performance statistics

According to that the CPU should do at least SSE. I don't know if the FreeBSD Version FreeNAS is based on already supported those CPUs (and the rest of the motherboard) specifically well, but I guess FreeBSD 10.(1) improved quite a bit. AFAIK those Atoms were released after FreeBSD 9.3, so maybe that explains, why it sucks a bit.

I also know that the pfsense guys "like" those motherboards, as they even use them for their own hardware. But pfsense is FreeBSD 10.1 based. If the board is still lingering around here, when FreeNAS 10 hits the ground, I might give it a second try. :)

Of course any other opinions/results are very welcome! I read much through this forums and I'm sure I will stumble upon future posts.

 

[Ericloewe]

The c2xx8 series lacks a lot of CPU instructions like SSE and others that are frequently used. I wouldn't be too surprised if performance overall just isn't what its cracked up to be. CPUs like the c2558 allow for a new Intel tech (forget the name) where purpose-built systems with purpose-built code that uses the new tech can see some amazing performance gains. But you won't see those gains with FreeNAS since it's a niche market for that kind of workload.

I'd recommend you go with the standard 2550 CPU instead of the 2558.

Of course, I don't know if anyone has provided any insight into how the 2558 performs in the past, so I can't tell you if my guess is right or not.

I think the only difference is fancy-networking-feature-with-incredibly-opaque-name on C2xx8 vs Turbo Boost on C2xx0.

 

[niaaard?]

Called "quickassist", FreeBSD support TBA. :)
(Yes that specific board is really marketed to networking appliances)

 

[cyberjock]

Called "quickassist", FreeBSD support TBA. :)
(Yes that specific board is really marketed to networking appliances)

That is it!