Windows 10 to be a free upgrade for Windows 7 and 8 users

[DanceswithUnix]

This hasn't been what I've observed, and what some benchmarks show for every day usage:

http://www.hecticgeek.com/2012/11/wi...ce-comparison/

Blimey that was hard work to follow. So, he measured writes in the two OS'es without re-installing them to the target device. When pointed out to him that the operating systems were running in different zones on the disk he re-ran the first test with big file copies (the easy one) to find that streaming writes are actually about the same. He reports the figures in an update, but leaves the misleading graph. He shows Windows being slower on small file copies despite having the benefit of twice the disk throughput from being in the outer HDD zone, which goes down to parity when he turns off the virus checker. Would have been nice if he had re-run that test in the same zone as well and got some meaningful figures (and bothered to update the graphs), but from where I am sitting I think that shows the usual small file win I expect to see from Linux.

Writing data to filesystems isn't that unusual. From the "smartctl -a" output on my workstation in the office:

241 Total_LBAs_Written 0x0000 100 253 000 Old_age Offline - 210222720796
242 Total_LBAs_Read 0x0000 100 253 000 Old_age Offline - 80879347223

I guess you write stuff, then tend to either overwrite it or read it back from cache.

[TheAnimus]

It's always going to be different use cases have different performance, the same way modern SSD controllers are almost impossible to benchmark without trying it with your use case. (a few years back I ended up having the intern spend a lot of time doing just that!).

But the fact is it's not half as bad as you make out. The biggest issue is that running windows without some on access virus scanner is sadly suicidal for most people. Meanwhile very few people exploit linux outside of the server world (where now all the pedants will point out that it isn't linux they are exploiting but some common component like perl or php).

My biggest performance bottleneck in my work stack isn't the OS at all, it's legacy ideas in how compilers work, which if you try to solve with a RAM drive often has a host of other issues.

[DanceswithUnix]

But the fact is it's not half as bad as you make out.

I can only go by my own experience, largely from writing & optimizing financial transaction systems across multiple platforms.

Still, I think I would rather use Windows than HPUX, it isn't *all* bad.

[TheAnimus]

I can only go by my own experience, largely from writing & optimizing financial transaction systems across multiple platforms.

Still, I think I would rather use Windows than HPUX, it isn't *all* bad.

Likewise, when SSDs first came out (2007), I found they had much better support for performance on Windows. The problem is, windows has more than one set of file APIs. One for normal everyday use, one for you are writing your own database server, you probably shouldn't be doing that grade performance.

This was more for doing financial modelling of the equity options market, I was doing some fun stuff with outperformance of index stuff. This is a classic geometric growth problem. So throwing more clusters at it gets expensive. I found that the performance of SSDs for essentially random access was so good, it enabled me to get rid of a half rack, with just one box.

I was doing all this in a nice managed language too (ie hacking new version all the time).

[crossy]

I can only go by my own experience, largely from writing & optimizing financial transaction systems across multiple platforms.

Still, I think I would rather use Windows than HPUX, it isn't *all* bad.

HPUX - all the in-box admin tools of Linux with the complexity of AIX. Personally I'd have prefered Linux's simplicity and AIX's admin tools. Still could be worse ... could be Solaris, and then you have to deal with Oracle (shudder).

What I'd like someone to explain is why running a McAfee scan on a native WindowsXP box brings said box to a near stop, yet running a McAfee scan on an XP VM running on top of Linux allows other work on that Linux system to continue, (albeit slower than normal). I really must get around to repeating that "test" on VMware ESX sometime...

[DanceswithUnix]

The problem is, windows has more than one set of file APIs. One for normal everyday use, one for you are writing your own database server, you probably shouldn't be doing that grade performance.

I would hope that when doing basic file access like open file, append a bit, close it again, then the everyday file API would be highly optimised for such everyday use

[shaithis]

What I'd like someone to explain is why running a McAfee scan on a native WindowsXP box brings said box to a near stop, yet running a McAfee scan on an XP VM running on top of Linux allows other work on that Linux system to continue, (albeit slower than normal). I really must get around to repeating that "test" on VMware ESX sometime...

The only thing that pops into my head is partition alignment. XP was the last desktop OS from Microsoft to not align properly. Wondering if the hypervisor is being clever and making the most of the "wasted" reads.

[TheAnimus]

I would hope that when doing basic file access like open file, append a bit, close it again, then the everyday file API would be highly optimised for such everyday use

To be honest you are sounding like the kind of programmer that drives the API team nuts, because they think of one use case, trivialise it in a manner that suggests any other use case is unlikely or completely compatible.

If you were a student of mine, I'd ask you to come back with 5 common use cases, that would apply to the standard home user that is your mother.

I'll give you a quick hint, are you limiting what's underneath that 'file stream'.

This is the problem with Windows, it is all things to all people.

[crossy]

The only thing that pops into my head is partition alignment. XP was the last desktop OS from Microsoft to not align properly. Wondering if the hypervisor is being clever and making the most of the "wasted" reads.

Oooh good thought - certainly better than mine, which was perhaps the Linux system was doing some kind of clever read-ahead caching.