Solaris 11 Struggles Against Linux Distributions

[Luke_Wolf]

Back to the OP... Solaris is getting the ass-end of any attention of Oracle. It is a shame to see that a once-great UNIX variant is languishing under Oracle, while they're off promoting a copy of CentOS as their flagship. Solaris carries the bloodline of UNIX, so unlike GNU/Linux it has some advantages in that it has avoided much of the bloat and feature creep of GNU/Linux, but I'd really like to see them either endorse illumos, or else reboot the OpenSolaris project. Sadly Oracle is headed by a dumbass, so I might as well sit back and take another bite of popcorn and sip my absinthe.

This thread has been dead for 2 years TeamBlackFox, let it be

[TeamBlackFox]

Back to the OP... Solaris is getting the ass-end of any attention of Oracle. It is a shame to see that a once-great UNIX variant is languishing under Oracle, while they're off promoting a copy of CentOS as their flagship. Solaris carries the bloodline of UNIX, so unlike GNU/Linux it has some advantages in that it has avoided much of the bloat and feature creep of GNU/Linux, but I'd really like to see them either endorse illumos, or else reboot the OpenSolaris project. Sadly Oracle is headed by a dumbass, so I might as well sit back and take another bite of popcorn and sip my absinthe.

[Luke_Wolf]

Phoronix should benchmark a Illumos based OS. Instead of the Oracle Solaris.

I'll also suggest a "TCP networking done right" - test against the network stacks of Linux and Illumos.

Oi... you. Stop Necroing old threads. If you want to whine about Linux's TCP/IP stack go make a thread for that, drudging up old threads is doing nothing more than inflating your post count.

[nasyt]

Phoronix should benchmark a Illumos based OS. Instead of the Oracle Solaris.

I'll also suggest a "TCP networking done right" - test against the network stacks of Linux and Illumos.

[unixconsole]

In a real world environment, you aren't going to change the compiler to make it faster. I very much doubt the benchmarks are "optimized" for linux. Although, I do agree with having some web benchmarks. This is also kind of an old article.

Yeah, but in the real world you'd use the recommended compiler for a given platform and gcc is not known for generating good code on non-Linux platforms. Solaris itself is not compiled with gcc either. And in if they were going to benchmark against a real BSD, gcc is not the recommended compiler either. Yeah, some web, java, and db benchmarks would be more realistic. The article says it was published on July 13, 2012...

[LinuxID10T]

1. You're using Solaris 11 Express and not Solaris 11 GA or even a patched version of Solaris 11. A lot has changed since the Express version and even more since the GA release. You should be testing this on the latest and greatest, otherwise you might as well go with older Linux distros as well.
2. If you're going to use GCC, you can install a newer version from OpenCSW or sunfreeware. Otherwise, you should use the native Solaris Studio compiler and optimize for 64bit.
3. For your libraries, you should do an "apples to apples" comparison which means using the same versions. Again, you can either look on OpenCSW or sunfreeware or you can compile newer versions.
4. Lastly, what is the validity of the benchmarks? Things like Himeno, SciMark, etc. don't exactly simulate real workloads or use cases. Also when you do NAS tests.. is it with NFSv3, NFSv4? If the benchmarks were designed and optimized for Linux, it's not exactly a good comparison. There are plenty of generic benchmarks like filebench, vdbench, iperf, etc. that are more meaningful. If you have the $$, the java and web benchmarks from SPEC are not bad either.

In a real world environment, you aren't going to change the compiler to make it faster. I very much doubt the benchmarks are "optimized" for linux. Although, I do agree with having some web benchmarks. This is also kind of an old article.

[unixconsole]

A few problems with this benchmarking..

1. You're using Solaris 11 Express and not Solaris 11 GA or even a patched version of Solaris 11. A lot has changed since the Express version and even more since the GA release. You should be testing this on the latest and greatest, otherwise you might as well go with older Linux distros as well.
2. If you're going to use GCC, you can install a newer version from OpenCSW or sunfreeware. Otherwise, you should use the native Solaris Studio compiler and optimize for 64bit.
3. For your libraries, you should do an "apples to apples" comparison which means using the same versions. Again, you can either look on OpenCSW or sunfreeware or you can compile newer versions.
4. Lastly, what is the validity of the benchmarks? Things like Himeno, SciMark, etc. don't exactly simulate real workloads or use cases. Also when you do NAS tests.. is it with NFSv3, NFSv4? If the benchmarks were designed and optimized for Linux, it's not exactly a good comparison. There are plenty of generic benchmarks like filebench, vdbench, iperf, etc. that are more meaningful. If you have the $$, the java and web benchmarks from SPEC are not bad either.

[vik1]

Yes, I was referring to the checksum functionality which is similar to having a software RAID 5 equivalent (but only one disk).

[Ansla]

@jadrevenge: no, the tests performed were not the default! Fedora uses BTRFS as the default and they changed it to use ext4. Anytime you use software RAID you incur performance penalties; that is well known and should be a giant caveat at the top of this article.

For an accurate comparison, do one of the following:
1) Solaris: ZFS boot partition and UFS for all others vs. Linux ext4 all partitions
2) Solaris: ZFS all partitions vs. Linux btrfs all partitions.

What "software RAID" are you talking about? RAID only makes sense when there are multiple phisical disks involved, why would ZFS use RAID on a single disk? Maybe you meant checksums?

[liam]

@jadrevenge: no, the tests performed were not the default! Fedora uses BTRFS as the default and they changed it to use ext4. Anytime you use software RAID you incur performance penalties; that is well known and should be a giant caveat at the top of this article.

For an accurate comparison, do one of the following:
1) Solaris: ZFS boot partition and UFS for all others vs. Linux ext4 all partitions
2) Solaris: ZFS all partitions vs. Linux btrfs all partitions.

Fedora doesn't use btrfs by default, nor will they until btrfs' repair tool is more general.