• Post Reply Bookmark Topic Watch Topic
  • New Topic

Speed comparisons  RSS feed

 
Frank Carver
Sheriff
Posts: 6920
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Here's something that is puzzling me. I use and maintain (amongst others) the two following systems:
'alex' SparcStation LX (I think that's a 45MHz Sparc), 48MB RAM, 3x400MB Sun SCSI HD, Solaris 2.6
'aylesbury' Gateway2000 Pentium 133, 48MB RAM, 1.6G EIDE HD, SuSE Linux 5.2
I have recently installed a parallel set of software on each machine so I can test out some server-side Java. Both machines already had GCC and Java 1.1.7, and I installed Apache 1.3.4 and Jserv 1.0b2 from sources. I then installed the same config files for both items and a set of my own Java classes.
Now for the speed comparisons. All the Sun literature I have seen implies that mere CPU speeds should not be used as a guide, as Sparc CPUs are efficient, and Sun hardware is optimised for network processing. So I wasn't expecting to see much diference.
On compilation, 'aylesbury' was a little faster, say 150%, but on running my tests I was astonished. 'aylesbury' was
consistently between 5 and 10 times faster in response and processing time for Java servlets!
Does anyone have any ideas why this might be? Admittedly they are on different networks, but both are 10Mb/s UTP.
is Sparc hardware really that much slower? Is Solaris really
that cumbersome?
The answers to this could significantly affect my future recommendations, so all comments welcome.
Frank.
------------------
--
Frank Carver
[ Personal: frank@efsol.com http://www.io.com/~efficacy ]
[ At Work: frank.carver@bt.com tel +44 (0)1473 227371 ]
 
Tim Uckun
Ranch Hand
Posts: 88
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
I wonder if it's because Apache was designed for Linux. Maybe it uses some feature of Linux that Solaris does not have. The only other clue may the JVM itself. Have you tried running an Java app outside of the Apache/Jrun environment. This would give you a clue as to the speed of the VM.
 
JD Childs
Greenhorn
Posts: 8
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
The LX is a dog...no doubt about it. I think your results have more to do with the speed of the BUS and "transactions" between RAM/Disk than anything else. There are tools you can download, like ByteBench, Perfmon, and IOBench, which can test the individual parts of your machine to determine where the bottleneck lies. It may also be that Solaris 2.6 loads a lot of crap in memory by default, whereas Linux may have many of those services turned off. For instance, Solaris may be shipped by default with an SMP-capable kernel, or RAID-aware drivers...Linux is not.
 
Matt Midcap
Sheriff
Posts: 440
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
I think JD is right about the "transactions" part, i.e. seeks, searches, etc.. You have three hard drives, although they're SCSI, you probably have things spread out quit a bit since they're a little on the small side. I wonder if you put everything on a 3GB SCSI, how much performace would increase (if any).
Course I could be wrong (I'm used to it, that's how I learn).
 
Frank Carver
Sheriff
Posts: 6920
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
And guess what, this morning I recieved my copy of "java Developers Journal" and one of the editorial pages ("straight talking", p14) is ranting about how expensive and inefficient Sun hardware is compared with generic Linux boxes. The author reckons he tried out a monster Enterprise 250 machine from sun at $6000, and found it was out performed by a Dell running Linux costing a third as much.
I'm definitely goiung to follow this up. If I can find some RAM for it, I've got another Sparc LX undaer a table in the machine room, and I'll install Linux on that. I'll also try solaris on an intel-based box and compare them too.
I know Linux is free, robust, flexible and well supported, but I didn't realize that it is probably *faster* than Solaris too!
------------------
--
Frank Carver
[ Personal: frank@efsol.com http://www.io.com/~efficacy ]
[ At Work: frank.carver@bt.com tel +44 (0)1473 227371 ]
 
Matt Midcap
Sheriff
Posts: 440
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Please let us know what your findings are! We are getting a new Sun box this year and I'll have the opportunity to play with it before it's needed for development.
I currently have a HP9000/700 (HP-UX), a HP 200MMX desktop running Linux, another HP200MMX desktop running SCO, and an Silicon Graphics Indy (150MX) in my cubical.
I'll run some performance tests with some Java apps and let you know the results. I'll also get all the configuration details of each box and post them too (for fair comparison sake).
Anyway, so far as well that I can see, the Linux box is just as fast as all the other machines running the same apps.
 
George Brown
Ranch Hand
Posts: 919
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Frank and Matt,
I had to resurrect this thread, it was just too tantalising. Did you get round to running those tests - were they conclusive? If you did, I'd love to find out what the results were.
 
Frank Carver
Sheriff
Posts: 6920
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Wow, I'd forgotten all about this thread. You really are digging around. I never got much further with this as "alex" trashed its primary HD soon after and I never found the time to rebuild it. I handed it over to a colleague when I left that contract. "Aylesbury" is still around here somewhere, although it hasn't been turned on since I reorganized my office a few months ago.
The whole thread is strangely relevent though, I'm just embarking on a "dot-com" venture of my own and have been struggling with what machine spec to buy for some potentially heavy server-side Java. In the end I've plumped for two identical PIII733,256MB,IBM30Gx7200-IDE 1U-rackmount boxes running RedHat 7 which I plan to configure up then send to a colo facility somewhere in the USA. They are both much more powerful than anything I've had control of before, but I've got no real idea how they will perform.
At the same time, my "real job" is currently working with some hugely overpriced (IMHO) Compaq/Dec alpha boxes. I haven't got enough control over them to set up software for speed comparisons, but they "feel" sluggish even compared to the Sparc 5 machines I used at a previous contract.
The hardware changes, but the issues remain the same ...
 
HS Thomas
Ranch Hand
Posts: 3404
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Now for the speed comparisons. All the Sun literature I have seen implies that mere CPU speeds should not be used as a guide, as Sparc CPUs are efficient, and Sun hardware is optimised for network processing. So I wasn't expecting to see much diference.

Would processing speeds of 300MHz and 600MHz be too slow for new Linux distributions ?
[ April 03, 2004: Message edited by: HS Thomas ]
 
Arjun Shastry
Ranch Hand
Posts: 1906
1
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Originally posted by HS Thomas:

Would processing speeds of 300MHz and 600MHz be too slow for new Linux distributions ?
[ April 03, 2004: Message edited by: HS Thomas ]

I don't think.I use sometimes 600 MHZ and Red Hat Enterprise edition is installed.If you work from command line,its good.GNOME,KDE are littlebit slower[much slower than windows].
 
Tim Holloway
Saloon Keeper
Posts: 18800
74
Android Eclipse IDE Linux
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
I believe that the baseline recommendation for Red Hat recent releases was something like 400MHz if running X and maybe P-233 if command-line only. Also I clocked an idle RH9 machine running X at about 104 MB RAM.
For Java, my general rule is AT LEAST 600MHz on the Intel platform, but I have one machine that runs on a P-200. Just takes 5 minutes to bring up JBoss, compared to under 30 sec on my Athlon 1800+.
Of course, having only 96MB RAM isn't making that poor P-233's job any easier!
 
Layne Lund
Ranch Hand
Posts: 3061
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
I'm running RH8 on a P166 with 64MB of RAM. Command line is just fine, but X is VERY sluggish. For the most part, I am simply using this box for a NFS server, so it works fine for me.
 
  • Post Reply Bookmark Topic Watch Topic
  • New Topic
Boost this thread!