Re: IDS 7.30 on Linux Tuning Benchmark?

> Very cool. I've been over in XPS too long. The high-performance loader > does this automagically... and there's no ontape for XPS just <gasp> onfubar... Given the military translation of fubar, I would cringe at the thought of trusting my recovery strategy to a product named onfubar. Mark Collins mcollins@us.dhl.com

Mark Collins wrote: > > > Very cool. I've been over in XPS too long. The high-performance loader > > does this automagically... and there's no ontape for XPS just <gasp> onfubar... > > Given the military translation of fubar, I would cringe at the thought of trusting > my recovery strategy to a product named onfubar. > > > > Mark Collins > mcollins@us.dhl.com Mark, I am hopeful that a new backup tool will come soon, as onbar, which I've nicknamed "onfubar", is woefully inadequate as a serious backup tool. It's one thing to sit in Silicon Valley night after night, thinking that the toolset is adequate, and another thing to actually try to use the tool. It sucks. When somebody at Informix wakes up, drinks some strong coffee, and really--no, REALLY--thinks about how badly onbar is designed, they will discover how bad it really is. Or maybe a better word, not how badly designed, but rather UNdesigned onbar really is. Anyway... Meanwhile, back on the loading thang... I went ahead with the ontape to NOLOGGING route, and loaded a table with 75 MB of data in 3 minutes 3 secs, row size 62 bytes, and the tuning parameters you and others here in this thread provided. This breaks out to roughly 16000 recs/sec, with 1,552,556 records in the loadfile. ( 1552556/93 = 16694 records per second ) I'd like to think after working this, that the engine is probably tuned for what it's going to really be doing, some really lightweight OLTP. I got a few foreground writes, but in a lot of ways, is this really bad? All the rows loaded, and nobody noticed, so what's the big deal? :-) Loading data is cyclical to a few minutes--a small fraction of the time normally spent on OLTP--that I wonder if it really needs to be better than this. After the index builds were done, the queries needed were coming back certainly faster than the SE system the customer is currently enjoying. I get the feeling from this cheaply PC solution that IDS on Linux is actually robust enough for a relatively serious shop. A small contingent of say 25 users banging on this server wouldn't really put a dent in the performance, especially using 4GL apps. Its really not a bad way to go... It seems to be able to handle quite a bit, even for cooked files. Any thoughts? Thanks, Tim -- - -- --- Tim Schaefer ---- tschaefe@mindspring.com --- http://www.inxutil.com -- -

Ooops ... Correction on my numbers for the record. I was chatting on the phone before and did math at the same time. :-) recs/secs 1552556/183 = 8484 records per second. I had reported previously for 1.5 minutes instead of 3 minutes 3 seconds. Still, not too bad for a small system is it? Tim -- - -- --- Tim Schaefer ---- tschaefe@mindspring.com --- http://www.inxutil.com -- -

Tim Schaefer wrote: [SNIP] > some really lightweight OLTP. I got a few foreground writes, but in a lot of ways, > is this really bad? All the rows loaded, and nobody noticed, so what's the big deal? :-) A few (<100 say) FG writes during data loading is no bug deal. It means that you might have tweaked a second or two more out the runtime if you had more buffers or fewer LRUS that's all. If you see it during production load only then you need to add buffers to the server. In this case call it unusual activity. Art S. Kagel

Thanks Art, This is the kind of system that will get some small, teeny tiny batch loads in the future, but the initial data load is what was "large". ( Compared to XPS loads it's nothing. ) Weekly cycles will probably be less than 1MB of data per load. Overall, I have to say, after working with the 7.30 version on Linux, it's a great product. I'd still like to know how it compares to the NT version and if NT is slower/faster. But what I'm really curious about is, when will the remaining features that are missing from the 7.30 port to Linux, be ported? Thanks, Tim "Art S. Kagel" wrote: > > Tim Schaefer wrote: > [SNIP] > > some really lightweight OLTP. I got a few foreground writes, but in a lot of ways, > > is this really bad? All the rows loaded, and nobody noticed, so what's the big deal? :-) > > A few (<100 say) FG writes during data loading is no bug deal. It > means that you might have tweaked a second or two more out the runtime > if you had more buffers or fewer LRUS that's all. If you see it during > production load only then you need to add buffers to the server. In > this case call it unusual activity. > > Art S. Kagel -- - -- --- Tim Schaefer ---- tschaefe@mindspring.com --- http://www.inxutil.com -- -