What do you use?

[Jason Bacon]

On Thursday 01 January 2004 07:09 am, Francisco Reyes wrote:
> On Thu, 1 Jan 2004, Scott Mitchell wrote:
> > As for RAID, we use Vinum, but only because I inherited a bunch of
> > machines with hot-swap SCSI bays and no hardware RAID.  It works well,
> > once you have it set up, and I've even managed to swap out failed drives
> > without a reboot
> >
> > :-)  I'll definitely investigate the 3ware cards when I need to build a
> > : new
> >
> > RAID server, though.
>
> But wouldn't a 3ware RAID be slower than an SCSI setup? Unless your
> current setup is using old SCSI disks. Also how is the load? Lots of
> simultaneous use or just many quick/small access (ie people using
> documents/spreadsheets).

Well, I'm in a position to provide some comparison data with not-too-many 
variables, so I felt obliged to run a quick benchmark.  The program writes a 
large file (large enough to overwhelm any memory buffering) using low-level 
I/O, does an fsync() before closing, and then tests random seek and 
sequential read on the same file.  I watched the system load using top during 
the benchmark and took a snapshot near the beginning and end of the write 
cycle.

~~~~~

3ware Escalade IDE 560gig RAID-5 (8 x 80G barracuda disks):

Writing 1677721600 byte test file...
Done.  Time = 58.433367
Performing 1000 random seeks and reads...
Done.  Time = 0.235705
Performing sequential read...
Done.  Time = 7.303429

CPU states:  0.8% user,  0.0% nice, 17.1% system,  0.0% interrupt, 82.2% idle
Mem: 68M Active, 609M Inact, 279M Wired, 89M Cache, 112M Buf, 960M Free
Swap: 4081M Total, 240K Used, 4081M Free

CPU states:  0.4% user,  0.0% nice, 13.2% system,  0.0% interrupt, 86.4% idle
Mem: 68M Active, 1390M Inact, 279M Wired, 88M Cache, 112M Buf, 180M Free
Swap: 4081M Total, 240K Used, 4081M Free

~~~~~

ICP Vortex SCA 240gig RAID-5 (4 x 72 gig cheetah SCA disks):

Writing 1677721600 byte test file...
Done.  Time = 53.667167
Performing 1000 random seeks and reads...
Done.  Time = 0.220799
Performing sequential read...
Done.  Time = 5.114555

CPU states:  0.4% user,  0.0% nice,  9.6% system,  1.0% interrupt, 89.1% idle
Mem: 52M Active, 401M Inact, 274M Wired, 77M Cache, 112M Buf, 1201M Free
Swap: 4081M Total, 248K Used, 4081M Free

CPU states:  0.4% user,  0.0% nice, 25.1% system,  1.9% interrupt, 72.6% idle
Mem: 52M Active, 1601M Inact, 276M Wired, 74M Cache, 112M Buf, 3228K Free
Swap: 4081M Total, 248K Used, 4081M Free

Both systems are running 5.1-RELEASE, with 2 gig RAM.  The 3ware machine has a 
faster CPU (Athlon 1.6Ghz) than the ICP (dual Pentium 850), but this 
shouldn't be significant for raw I/O at these CPU speeds.

The CPU usage is lower on the ICP system (remember that it has a slower CPU 
when comparing the CPU loads).  CPU use consistently went up toward the end 
of the write cycle on the ICP system, but stayed flat on the 3ware.  

I noticed a drastic reduction of CPU load on the 3ware system when we went 
from 4.5-RELEASE to 4.6-RELEASE  CPU load has never been a problem on either 
server.

	JB