Re: Best database for web backend on Linux?
This is a multi-part message in MIME format. ------=_NextPart_000_1792_01BDFC67.5B5FFBD0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Interesting. I've done lots of Sybase work, and found I could tune = Oracle to be faster than Sybase in everything except for loading a = clustered index in reverse order where the data was already pre-sorted. = If left to a random distribution it was faster than Sybase. Also if you = push sybase for extended periods of time, their delayed mechanism really = starts to hurt you. For accounting applications I found the dirty read to be a great problem = when doing trial balances. I also found the math module in Sybase to SUCK! I mean we got different = answers when running the exact same stored procedure on DEC VMS, UNIX, = etc. It uses the host mathlibs, which on VMS aren't all that hot. Now if you take the benchmark that "Corporate Computing" does, you'll = see that Oracle still kicks ass. But you can always slow down anything, = if you try hard enough. Corporate computing takes an application from a real business, loads = it onto any rdbms with guts to enter the contest, and then runs typical = queries, indexing, loading, inserts, etc, that the business does in a = regular work year. Now that's a real bench mark. Sun has a cool benchmark, as does = Qualcomm, and Oracle always does well in those environments. REGARDLESS, having to write 40% less code, because the rdbms does = row-level locking for you is worth it to me. Thor HW Evan Carew wrote in message <362CE8BE.53F4CDB0@yahoo.com>... Good question Thor. It is important to read those benchmarks very = carefully. When Sybase and Oracle came to my job site to do a bake off, = we noticed they were utilizing a special setting for their server = designed for the benchmarks. This setting of course disables row level = locking. When we forced both vendors to do "out of box" installs the = picture changed substantially. When we also did feature by feature = comparison, it was interesting to note just how badly you could make = Oracle perform. As a programmer somewhat familliar with the history of = the two products this didn't suprise me overly much. Sybase's history = includes being the first fully client server database whereas Oracle's = includes being a reasonable mainframe based database that converted to = the client server model (some would say not very well).=20 Should you ever have the chance, you want to force a vendor to do = thourough stored proceeduretesting while someone runs a large query = where the result set is guaranteed to be larger than the memory pool = alloted to the data server. You also want to be shure you review all the = database performance settings so they don't manage to pull the wool over = your eyes (you would hope this would go without saying). The last time I = did this, it managed to crash the INFORMIX box (a bug in their = implementation of stored procedures) and really sloooow down the oracle = box. The sybase server by comparison didn't sag too much. You also want = to get a feeling for what feature aren't shared by all the servers, for = instance, Sybase has the ability to partition its memory pool so the = neat trick mentioned above will have no effect on performance. Another = issue is that there are certain SQL dvanced optimizations that are = impossible with ORACLE . For refrences to this, see Optimizing = Transact-SQL Programming Techniques by Dr. David Rozeshtein & Anatoly = Abramovich .=20 =20 Thor wrote:=20 =20 I like row-level locking. Everyone is trying to get it into their = dbms. It sure saves a tonne of coding (grab the sybase docs and read = the timestamp work they suggest for high availability and avoid = blocking, yuck).If row-level locking is implemented well, it doesn't = need to grind the database to a standstill, maybe if it's an = afterthought/option to add, you might see that.If that were the case, = then why is Oracle able to post such fast benchmarks? Thor HW =20 --=20 ----=3D=3D-- _ / / \\ ---=3D=3D---(_)__ __ ____ __ / / /\\ \\ --=3D=3D---/ / _ \\/ // /\\ \\/ / / /_/\\ \\ \\ -=3D=3D=3D=3D=3D/_/_//_/\\_,_/ /_/\\_\\ /______\\ \\ \\ A proud member of TeamLinux \\_________\\/ email: carew@home.msen.com web: http://home.msen.com/~carew =20 ------=_NextPart_000_1792_01BDFC67.5B5FFBD0 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable <!DOCTYPE HTML PUBLIC "-//W3C//DTD W3 HTML//EN"> <HTML> <HEAD> <META content=3Dtext/html;charset=3Diso-8859-1 = http-equiv=3DContent-Type> <META content=3D'"MSHTML 4.71.1712.3"' name=3DGENERATOR> </HEAD> <BODY bgColor=3D#ffffff> <DIV><FONT color=3D#000000 size=3D2>Interesting. I've done lots of = Sybase=20 work, and found I could tune Oracle to be faster than Sybase in = everything=20 except for loading a clustered index in reverse order where the data was = already=20 pre-sorted. If left to a random distribution it was faster than=20 Sybase. Also if you push sybase for extended periods of time, = their=20 delayed mechanism really starts to hurt you.</FONT></DIV> <DIV><FONT color=3D#000000 size=3D2></FONT> </DIV> <DIV><FONT size=3D2>For accounting applications I found the dirty read = to be a=20 great problem when doing trial balances.</FONT></DIV> <DIV><FONT size=3D2></FONT> </DIV> <DIV><FONT size=3D2>I also found the math module in Sybase to = SUCK! I mean=20 we got different answers when running the exact same stored procedure on = DEC=20 VMS, UNIX, etc. It uses the host mathlibs, which on VMS aren't all = that=20 hot.</FONT></DIV> <DIV><FONT size=3D2></FONT> </DIV> <DIV><FONT color=3D#000000 size=3D2>Now if you take the benchmark that=20 "Corporate Computing" does, you'll see that Oracle still kicks = ass. But you can always slow down anything, if you try hard=20 enough.</FONT></DIV> <DIV><FONT color=3D#000000 size=3D2> Corporate = computing takes an=20 application from a real business, loads it onto any rdbms with guts to = enter the=20 contest, and then runs typical queries, indexing, loading, inserts, etc, = that=20 the business does in a regular work year.</FONT> </DIV> <DIV><FONT color=3D#000000 size=3D2>Now that's a real bench mark. = Sun has a=20 cool benchmark, as does Qualcomm, and Oracle always does well in those=20 environments.</FONT></DIV> <DIV><FONT color=3D#000000 size=3D2></FONT> </DIV> <DIV><FONT size=3D2>REGARDLESS, having to write 40% less code, because = the rdbms=20 does row-level locking for you is worth it to me.</FONT></DIV> <DIV><FONT size=3D2></FONT> </DIV> <DIV><FONT color=3D#000000 size=3D2>Thor HW</FONT></DIV> <BLOCKQUOTE=20 style=3D"BORDER-LEFT: #000000 solid 2px; MARGIN-LEFT: 5px; PADDING-LEFT: = 5px"> <DIV>Evan Carew<ECAREW@YAHOO.COM> wrote in message <<A=20 = href=3D"mailto:362CE8BE.53F4CDB0@yahoo.com">362CE8BE.53F4CDB0@yahoo.com</= A>>...</DIV>Good=20 questi