RE: ejb performance
Yes, of course: Pay attention to use write location in your code, f.ex. I use jdbc/CBOSS to locate the datasource since it is the place where pooled datasources are stored by orion. Regards, Savotchkin Egor Web-technologies department eMail: [EMAIL PROTECTED] CJSC "SoftPro" - Billing systems tel: (+7-095) 755-5655 (3135) web-site: http://www.cboss.ru -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Marcus Lankenau Sent: Saturday, December 16, 2000 6:30 PM To: Orion-Interest Subject: Re: ejb performance Hi Savotchkin! Could you please describe, what you did to activate pooling? I tried this for my application, but without success... (could you maybe post you datasource xml?) thx in advance Marcus Lankenau > Yeah!!! I forgot that in this function I got not pooled db connection. So, > after I turned the pooling on it became 0 to 10 ms (on PIII 600 with 256 > RAM) to invoke this function!!! Great indexes!! > So given these numbers I may throw away all these transaction attributes > and f.ex. always specify requires, since it doesnot affect perfomance... > > Regards, >Savotchkin Egor > > Web-technologies department > eMail: [EMAIL PROTECTED] > CJSC "SoftPro" - Billing systems > tel: (+7-095) 755-5655 (3135) > web-site: http://www.cboss.ru > -- begin:vcard n:Lankenau;Marcus x-mozilla-html:FALSE url:wwl.de org:wwl vision2market;Interactive adr:;;goebelstr. 46;Lilienthal;D;28865; version:2.1 email;internet:[EMAIL PROTECTED] title:Senior Developer fn:Marcus Lankenau end:vcard
Re: ejb performance
Hi Savotchkin! Could you please describe, what you did to activate pooling? I tried this for my application, but without success... (could you maybe post you datasource xml?) thx in advance Marcus Lankenau > Yeah!!! I forgot that in this function I got not pooled db connection. So, > after I turned the pooling on it became 0 to 10 ms (on PIII 600 with 256 > RAM) to invoke this function!!! Great indexes!! > So given these numbers I may throw away all these transaction attributes > and f.ex. always specify requires, since it doesnot affect perfomance... > > Regards, >Savotchkin Egor > > Web-technologies department > eMail: [EMAIL PROTECTED] > CJSC "SoftPro" - Billing systems > tel: (+7-095) 755-5655 (3135) > web-site: http://www.cboss.ru > -- begin:vcard n:Lankenau;Marcus x-mozilla-html:FALSE url:wwl.de org:wwl vision2market;Interactive adr:;;goebelstr. 46;Lilienthal;D;28865; version:2.1 email;internet:[EMAIL PROTECTED] title:Senior Developer fn:Marcus Lankenau end:vcard
RE: ejb performance
Yeah!!! I forgot that in this function I got not pooled db connection. So, after I turned the pooling on it became 0 to 10 ms (on PIII 600 with 256 RAM) to invoke this function!!! Great indexes!! So given these numbers I may throw away all these transaction attributes and f.ex. always specify requires, since it doesnot affect perfomance... Regards, Savotchkin Egor Web-technologies department eMail: [EMAIL PROTECTED] CJSC "SoftPro" - Billing systems tel: (+7-095) 755-5655 (3135) web-site: http://www.cboss.ru -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Robert Krueger Sent: Saturday, December 16, 2000 2:57 PM To: Orion-Interest Subject: Re: ejb performance At 14:28 16.12.00 , you wrote: >Hi all! > I made some tests to find out how transactions affect > performance. I called >a sessionless' ejb empty function with different transaction attributes. It >turned out that call time mean was about 250 ms and it almost not changed >depending on attributesn (Requires, supports, ...). So, does transactions >have so little impact on perfomance??? > Why it takes about 250ms to invoke an empty function on ejb? Is > it because >ejb are distrubuted objects? What else could affect the ejb performance? > >Regards, >Savotchkin Egor you must have made a mistake 250ms is ridiculously slow. that's 4 calls per sec please check your setup. this doesn't reflect our observations at all. even with transactions that's way too slow or are you by any chance using a 386 with 4 Meg RAM and a nojit VM ;-)). regards, robert >Web-technologies department >eMail: [EMAIL PROTECTED] >CJSC "SoftPro" - Billing systems >tel: (+7-095) 755-5655 (3135) >web-site: http://www.cboss.ru > (-) Robert Krüger (-) SIGNAL 7 Gesellschaft für Informationstechnologie mbH (-) Brüder-Knauß-Str. 79 - 64285 Darmstadt, (-) Tel: 06151 665401, Fax: 06151 665373 (-) [EMAIL PROTECTED], www.signal7.de
Re: ejb performance
At 14:28 16.12.00 , you wrote: >Hi all! > I made some tests to find out how transactions affect > performance. I called >a sessionless' ejb empty function with different transaction attributes. It >turned out that call time mean was about 250 ms and it almost not changed >depending on attributesn (Requires, supports, ...). So, does transactions >have so little impact on perfomance??? > Why it takes about 250ms to invoke an empty function on ejb? Is > it because >ejb are distrubuted objects? What else could affect the ejb performance? > >Regards, >Savotchkin Egor you must have made a mistake 250ms is ridiculously slow. that's 4 calls per sec please check your setup. this doesn't reflect our observations at all. even with transactions that's way too slow or are you by any chance using a 386 with 4 Meg RAM and a nojit VM ;-)). regards, robert >Web-technologies department >eMail: [EMAIL PROTECTED] >CJSC "SoftPro" - Billing systems >tel: (+7-095) 755-5655 (3135) >web-site: http://www.cboss.ru > (-) Robert Krüger (-) SIGNAL 7 Gesellschaft für Informationstechnologie mbH (-) Brüder-Knauß-Str. 79 - 64285 Darmstadt, (-) Tel: 06151 665401, Fax: 06151 665373 (-) [EMAIL PROTECTED], www.signal7.de
RE: EJB Performance Question.
At 10:11 AM 11/15/2000 +0100, you wrote: >At 14:48 14.11.00 , you wrote: > >>Thanks Robert. I think I'll try running some >>benchmarks this week and post the results. I wonder.. >>is there a way for and EJB->EJB to be _forced_ to >>go through RMI? i.e. can I turn this optimization >>off? > >I don't think there is a documented way to to this. Not to be a jerk but to provide some constructive criticism to Orion ... I don't think there is a documented way to do many, many things. Considering that it is a closed-source product, I think documentation should be one of the most important things to accompany the actual binaries.
RE: EJB Performance Question.
The idea behind a bulk accessor is to return a single
object that holds all the data of your EJB.
(Forgive me for any syntax mistakes. I'm doing this
from heart and I've done so much EJB2.0 stuff that my
1.1 is a little rusty.)
If you have an EJB like this:
public class FooEJB implements EntityBean
{
public String data1;
public String data2;
public int data3;
...
// this is a bulk accessor
public FooView getFoo()
{
return new FooView(data1, data2, data3);
}
}
You would create a "view" object that can hold all
the EJB data. A view is a very thin "bean-like"
object:
public class FooView
{
String data1;
String data2;
int data3;
public String getData1() { return data1; }
public String getData2() { return data2; }
public int getData3() { return data3; }
public FooView(String data1, String data2, int data3)
{
this.data1 = data1;
this.data2 = data2;
this.data3 = data3;
}
}
Now you client only need to call:
FooView fv = fooejb.getFooView();
to get all the data within the Foo EJB. This reduces
the RMI marshalling to a single call. Then, you use
the FooView accessors to get to the data; since FooView
is a local object, this will be fast.
NOTE: if you followed this thread, you heard from others
that Orion optimizes intra-VM EJB calls to "call-by-value"
which doesn't use RMI. This is great for Orion, but is
not specified in the spec, so if you were to port your
application to another container, you aren't guaranteed
this optimization. The above bulk accessor/View pattern
will always work.
-tim
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, November 14, 2000 7:04 PM
To: Orion-Interest
Subject: Re: EJB Performance Question.
Thank You for your kind reply. Can I get some pseudo-code an example?
Thank you.
- Original Message -
From: Tim Drury
To: Orion-Interest
Sent: Wednesday, November 15, 2000 12:03 AM
Subject: RE: EJB Performance Question.
Every single one of those calls to dir.getXXX() has to go across
the network via RMI. This is slow. You are better off using a
bulk accessor pattern. For example, create a new class called
DirView which contains all the attributes of your Dir EJB. Then
make a single call to the Dir EJB to get a DirView. This will cause
only 1 RMI call and save you a huge amount of time.
Do you understand? I can probably pseudo-code an example
if necessary.
-tim
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, November 14, 2000 3:56 AM
To: Orion-Interest
Subject: EJB Performance Question.
Hi, every one. First i'm sorry for my english.
We use servlet that call EJB. Next is code fraction.
//---
public Vector findByFirstPage(DirHome home, Integer rowCount) throws
Exception{
Vector rows = new Vector();
Dir dir = null;
System.out.println("step 11 time : " + (new java.util.Date()));
// call EJB
Collection col = home.findByFirstPage(rowCount.intValue());
Iterator iter = col.iterator();
System.out.println("step 12 time : " + (new java.util.Date()));
while(iter.hasNext()) {
dir = (Dir)iter.next();
rows.add(EJBToRow(dir)); //<--- bottle neck #
}
System.out.println("step 13 time : " + (new java.util.Date()));
return rows;
}
//---
Simple code. In while loop, EJBToRow() method take 3 second each call.
Next code is EJBToRow(). Very Simple. Only call EJB Meber methods.
//---
public Vector EJBToRow(Dir dir) throws Exception {
Vector row = new Vector();
System.out.println("step 21 time : " + (new java.util.Date()));
row.add(dir.getId());
row.add(new Long(dir.getPId()));
row.add(dir.getName());
row.add(new Long(dir.getSerial()));
row.add(new Long(dir.getChildCount()));
row.add(new Long(dir.getDepth()));
row.add(dir.getPMap());
row.add(dir.getType());
row.add(dir.getUserId());
row.add(dir.getGroupId());
row.add(dir.getOwnerPermR());
row.add(dir.getOwnerPermW());
row.add(dir.getOwnerPermX());
row.add(dir.getGroupPermR());
row.add(dir.getGroupPermW());
row.add(dir.getGroupPermX());
row.add(dir.getOtherPermR());
row.add(dir.getOtherPermW());
row.add(dir.getOtherPermX());
row.add(dir.getCreateDate());
row.add(dir.getUpdateDate());
row.add(dir.getExpireDate());
row.add(dir.getRemark());
System.out.println("step 22 time : " + (new java.util.Date()));
return row;
}
//---
What's the key problem?
e-mail
site: [EMAIL PROTECTED]
: www.javanuri.comÀüÈ
mobile : 02-6257-3002
: 019-255-2855
RE: EJB Performance Question.
At 14:48 14.11.00 , you wrote: >Thanks Robert. I think I'll try running some >benchmarks this week and post the results. I wonder.. >is there a way for and EJB->EJB to be _forced_ to >go through RMI? i.e. can I turn this optimization >off? I don't think there is a documented way to to this. robert >-tim > > > > -Original Message- > > From: Robert Krueger [mailto:[EMAIL PROTECTED]] > > Sent: Tuesday, November 14, 2000 1:36 PM > > To: Orion-Interest > > Subject: RE: EJB Performance Question. > > > > > > At 12:06 14.11.00 , you wrote: > > > > At 10:03 14.11.00 , you wrote: > > > > >Every single one of those calls to dir.getXXX() has to go across > > > > >the network via RMI. This is slow. You are better off using a > > > > > > > > well, with orion this is intra-vm so its not that bad ... > > > > > >Is there proof that Orion does this? I'm not trying to be a jerk, > > >but I'd like to see some numbers on how long a set/get takes for > > >a client-to-ejb call and an ejb-to-ejb (same vm) call. > > > > do a getClass().getName() on your ejbobject. will be the same > > wrapper class > > that implements the ejbobject (if you look into the generate > > code that's > > temporarily there on deployment). > > > > anyway, I've never benchmarked but it is orders of magnitude > > faster than > > any rmi stack I've seen. I think there's even an old mail in > > the archive > > from karl that confirms this otiomization because it was one > > of the first > > things I asked (must at least be 10 months ago now). > > > > regards, > > > > robert > > > > >Not even Weblogic optimizes to pass-by-value for intra-vm ejb > > > > > >calls. You have to use TopLink to get that optimization. > > >Funny though, Weblogic provides a tag in their deployment > > >description but it doesn't do anything. > > > > > >-tim > > > > (-) Robert Krüger > > (-) SIGNAL 7 Gesellschaft für Informationstechnologie mbH > > (-) Brüder-Knauß-Str. 79 - 64285 Darmstadt, > > (-) Tel: 06151 665401, Fax: 06151 665373 > > (-) [EMAIL PROTECTED], www.signal7.de > > > > (-) Robert Krüger (-) SIGNAL 7 Gesellschaft für Informationstechnologie mbH (-) Brüder-Knauß-Str. 79 - 64285 Darmstadt, (-) Tel: 06151 665401, Fax: 06151 665373 (-) [EMAIL PROTECTED], www.signal7.de
Re: EJB Performance Question.
Thank You for your kind reply. Can I get some pseudo-code an example?
Thank you.
- Original Message -
From: Tim Drury
To: Orion-Interest
Sent: Wednesday, November 15, 2000 12:03
AM
Subject: RE: EJB Performance Question.
Every single one of those calls to dir.getXXX() has to go
across
the
network via RMI. This is slow. You are better off using
a
bulk
accessor pattern. For example, create a new class
called
DirView which contains all the attributes of your Dir EJB.
Then
make
a single call to the Dir EJB to get a DirView. This will
cause
only
1 RMI call and save you a huge amount of time.
Do
you understand? I can probably pseudo-code an
example
if
necessary.
-tim
-Original Message-From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]Sent: Tuesday, November 14, 2000
3:56 AMTo: Orion-InterestSubject: EJB Performance
Question.
Hi, every one. First i'm sorry for my
english.
We use servlet that call EJB. Next is code
fraction.
//---
public Vector findByFirstPage(DirHome home, Integer
rowCount) throws Exception{ Vector
rows = new Vector(); Dir dir =
null;
System.out.println("step 11 time : " + (new
java.util.Date()));
// call
EJB Collection col =
home.findByFirstPage(rowCount.intValue()); Iterator iter =
col.iterator();
System.out.println("step 12 time : " + (new
java.util.Date()));
while(iter.hasNext()) {
dir = (Dir)iter.next(); rows.add(EJBToRow(dir));
//<--- bottle neck
# } System.out.println("step 13
time : " + (new java.util.Date()));
return
rows; }
//---
Simple code. In while loop, EJBToRow() method take 3
second each call.
Next code is EJBToRow(). Very Simple. Only call EJB Meber
methods.
//---
public Vector EJBToRow(Dir dir) throws Exception
{ Vector row = new Vector();
System.out.println("step 21 time : " + (new
java.util.Date()));
row.add(dir.getId()); row.add(new
Long(dir.getPId())); row.add(dir.getName()); row.add(new
Long(dir.getSerial())); row.add(new
Long(dir.getChildCount())); row.add(new
Long(dir.getDepth())); row.add(dir.getPMap()); row.add(dir.getType()); row.add(dir.getUserId()); row.add(dir.getGroupId()); row.add(dir.getOwnerPermR()); row.add(dir.getOwnerPermW()); row.add(dir.getOwnerPermX()); row.add(dir.getGroupPermR()); row.add(dir.getGroupPermW()); row.add(dir.getGroupPermX()); row.add(dir.getOtherPermR()); row.add(dir.getOtherPermW()); row.add(dir.getOtherPermX()); row.add(dir.getCreateDate()); row.add(dir.getUpdateDate()); row.add(dir.getExpireDate()); row.add(dir.getRemark());
System.out.println("step 22 time : " + (new
java.util.Date()));
return
row; }
//---
What's the key problem?
e-mail
site
: [EMAIL PROTECTED]
:
www.javanuri.com
ÀüÈ
mobile
: 02-6257-3002
: 019-255-2855
RE: EJB Performance Question.
Thanks Robert. I think I'll try running some benchmarks this week and post the results. I wonder.. is there a way for and EJB->EJB to be _forced_ to go through RMI? i.e. can I turn this optimization off? -tim > -Original Message- > From: Robert Krueger [mailto:[EMAIL PROTECTED]] > Sent: Tuesday, November 14, 2000 1:36 PM > To: Orion-Interest > Subject: RE: EJB Performance Question. > > > At 12:06 14.11.00 , you wrote: > > > At 10:03 14.11.00 , you wrote: > > > >Every single one of those calls to dir.getXXX() has to go across > > > >the network via RMI. This is slow. You are better off using a > > > > > > well, with orion this is intra-vm so its not that bad ... > > > >Is there proof that Orion does this? I'm not trying to be a jerk, > >but I'd like to see some numbers on how long a set/get takes for > >a client-to-ejb call and an ejb-to-ejb (same vm) call. > > do a getClass().getName() on your ejbobject. will be the same > wrapper class > that implements the ejbobject (if you look into the generate > code that's > temporarily there on deployment). > > anyway, I've never benchmarked but it is orders of magnitude > faster than > any rmi stack I've seen. I think there's even an old mail in > the archive > from karl that confirms this otiomization because it was one > of the first > things I asked (must at least be 10 months ago now). > > regards, > > robert > > >Not even Weblogic optimizes to pass-by-value for intra-vm ejb > > > >calls. You have to use TopLink to get that optimization. > >Funny though, Weblogic provides a tag in their deployment > >description but it doesn't do anything. > > > >-tim > > (-) Robert Krüger > (-) SIGNAL 7 Gesellschaft für Informationstechnologie mbH > (-) Brüder-Knauß-Str. 79 - 64285 Darmstadt, > (-) Tel: 06151 665401, Fax: 06151 665373 > (-) [EMAIL PROTECTED], www.signal7.de > >
RE: EJB Performance Question.
If you're referring to the enable-call-by-reference tag, then indeed it does do something (at lease in version 5.1). Setting this tag to true will effectively pass EJB method arguments and returned objects by reference when called within the same JVM, rather than by value in accordance with the EJB 1.1 specification. This obviously increases performance by eliminating the marshalling overhead, at the cost of being non-spec compliant. At any rate, for better or worse, in WebLogic call by reference is enabled by default. In other words, calls between servlets and EJBs in the same JVM are optimized. That said, don't let these optimizations pollute your EJB design. Having a well-defined and minimal remote contract is much more portable, and generally easier to maintain and manage. Moreover, you may at some point in the future decide to separate the web server and EJB container into separate JVMs. In general, EJB interface contracts are specifically important to design correctly, as the invocation of remote methods may in turn provoke life cycle methods, transactions, etc. Mike --- Tim Drury <[EMAIL PROTECTED]> wrote: > > At 10:03 14.11.00 , you wrote: > > >Every single one of those calls to dir.getXXX() has to go across > > >the network via RMI. This is slow. You are better off using a > > > > well, with orion this is intra-vm so its not that bad ... > > Is there proof that Orion does this? I'm not trying to be a jerk, > but I'd like to see some numbers on how long a set/get takes for > a client-to-ejb call and an ejb-to-ejb (same vm) call. > > Not even Weblogic optimizes to pass-by-value for intra-vm ejb > calls. You have to use TopLink to get that optimization. > Funny though, Weblogic provides a tag in their deployment > description but it doesn't do anything. > > -tim > > > > __ Do You Yahoo!? Yahoo! Calendar - Get organized for the holidays! http://calendar.yahoo.com/
RE: EJB Performance Question.
At 12:06 14.11.00 , you wrote: > > At 10:03 14.11.00 , you wrote: > > >Every single one of those calls to dir.getXXX() has to go across > > >the network via RMI. This is slow. You are better off using a > > > > well, with orion this is intra-vm so its not that bad ... > >Is there proof that Orion does this? I'm not trying to be a jerk, >but I'd like to see some numbers on how long a set/get takes for >a client-to-ejb call and an ejb-to-ejb (same vm) call. do a getClass().getName() on your ejbobject. will be the same wrapper class that implements the ejbobject (if you look into the generate code that's temporarily there on deployment). anyway, I've never benchmarked but it is orders of magnitude faster than any rmi stack I've seen. I think there's even an old mail in the archive from karl that confirms this otiomization because it was one of the first things I asked (must at least be 10 months ago now). regards, robert >Not even Weblogic optimizes to pass-by-value for intra-vm ejb > >calls. You have to use TopLink to get that optimization. >Funny though, Weblogic provides a tag in their deployment >description but it doesn't do anything. > >-tim (-) Robert Krüger (-) SIGNAL 7 Gesellschaft für Informationstechnologie mbH (-) Brüder-Knauß-Str. 79 - 64285 Darmstadt, (-) Tel: 06151 665401, Fax: 06151 665373 (-) [EMAIL PROTECTED], www.signal7.de
RE: EJB Performance Question.
> At 10:03 14.11.00 , you wrote: > >Every single one of those calls to dir.getXXX() has to go across > >the network via RMI. This is slow. You are better off using a > > well, with orion this is intra-vm so its not that bad ... Is there proof that Orion does this? I'm not trying to be a jerk, but I'd like to see some numbers on how long a set/get takes for a client-to-ejb call and an ejb-to-ejb (same vm) call. Not even Weblogic optimizes to pass-by-value for intra-vm ejb calls. You have to use TopLink to get that optimization. Funny though, Weblogic provides a tag in their deployment description but it doesn't do anything. -tim
RE: EJB Performance Question.
At 10:03 14.11.00 , you wrote:
>Every single one of those calls to dir.getXXX() has to go across
>the network via RMI. This is slow. You are better off using a
well, with orion this is intra-vm so its not that bad ...
>bulk accessor pattern. For example, create a new class called
.. although a bulk accessor is generally a good thing to use in such
situations.
have you tried wrapping one usertransaction around your update operation or
maybe even around the entire loop, depending on what you want to see as a
transaction. this will improve performance significantly because you don't
have all the tx overhead. if you don't know what a usertransaction is, read
up on JTA.
HTH
robert
>DirView which contains all the attributes of your Dir EJB. Then
>make a single call to the Dir EJB to get a DirView. This will cause
>only 1 RMI call and save you a huge amount of time.
>
>Do you understand? I can probably pseudo-code an example
>if necessary.
>
>-tim
>
>
>>-Original Message-
>>From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
>>Sent: Tuesday, November 14, 2000 3:56 AM
>>To: Orion-Interest
>>Subject: EJB Performance Question.
>>
>>Hi, every one. First i'm sorry for my english.
>>
>>We use servlet that call EJB. Next is code fraction.
>>
>>//---
>>public Vector findByFirstPage(DirHome home, Integer rowCount) throws
>>Exception{
>>
>> Vector rows = new Vector();
>> Dir dir = null;
>>
>> System.out.println("step 11 time : " + (new java.util.Date()));
>>
>> // call EJB
>> Collection col = home.findByFirstPage(rowCount.intValue());
>> Iterator iter = col.iterator();
>>
>> System.out.println("step 12 time : " + (new java.util.Date()));
>>
>> while(iter.hasNext()) {
>> dir = (Dir)iter.next();
>> rows.add(EJBToRow(dir)); //<--- bottle neck #
>> }
>>
>> System.out.println("step 13 time : " + (new java.util.Date()));
>>
>> return rows;
>>
>> }
>> //---
>>
>>
>>Simple code. In while loop, EJBToRow() method take 3 second each call.
>>
>>Next code is EJBToRow(). Very Simple. Only call EJB Meber methods.
>>
>> //---
>> public Vector EJBToRow(Dir dir) throws Exception {
>>
>> Vector row = new Vector();
>>
>>System.out.println("step 21 time : " + (new java.util.Date()));
>>
>> row.add(dir.getId());
>> row.add(new Long(dir.getPId()));
>> row.add(dir.getName());
>> row.add(new Long(dir.getSerial()));
>> row.add(new Long(dir.getChildCount()));
>> row.add(new Long(dir.getDepth()));
>> row.add(dir.getPMap());
>> row.add(dir.getType());
>> row.add(dir.getUserId());
>> row.add(dir.getGroupId());
>> row.add(dir.getOwnerPermR());
>> row.add(dir.getOwnerPermW());
>> row.add(dir.getOwnerPermX());
>> row.add(dir.getGroupPermR());
>> row.add(dir.getGroupPermW());
>> row.add(dir.getGroupPermX());
>> row.add(dir.getOtherPermR());
>> row.add(dir.getOtherPermW());
>> row.add(dir.getOtherPermX());
>> row.add(dir.getCreateDate());
>> row.add(dir.getUpdateDate());
>> row.add(dir.getExpireDate());
>> row.add(dir.getRemark());
>>
>>System.out.println("step 22 time : " + (new java.util.Date()));
>>
>> return row;
>>
>> }
>> //---
>>
>>What's the key problem?
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>--
>>e-mail
>>site
>>: [EMAIL PROTECTED]
>>: www.javanuri.com
>>ÀüÈ
>>mobile
>>: 02-6257-3002
>>: 019-255-2855
>>
>>
>>
>>--
>>
(-) Robert Krüger
(-) SIGNAL 7 Gesellschaft für Informationstechnologie mbH
(-) Brüder-Knauß-Str. 79 - 64285 Darmstadt,
(-) Tel: 06151 665401, Fax: 06151 665373
(-) [EMAIL PROTECTED], www.signal7.de
Re: EJB Performance Question.
Hi ±èöȸ,
> public Vector EJBToRow(Dir dir) throws Exception {
[...]
> row.add(dir.getId());
> row.add(new Long(dir.getPId()));
> row.add(dir.getName());
> row.add(new Long(dir.getSerial()));
> row.add(new Long(dir.getChildCount()));
> row.add(new Long(dir.getDepth()));
[...]
> What's the key problem?
thats not a problem, thats a feature!
(each remote call runs within an transaction...)
study all the EJB-docs you can get and redesign your app.
bye
klaus
--
Klaus Thiele - Personal & Informatik AG
mailto:[EMAIL PROTECTED]
"There's got to be more to life than compile-and-go."
RE: EJB Performance Question.
Every
single one of those calls to dir.getXXX() has to go across
the
network via RMI. This is slow. You are better off using
a
bulk
accessor pattern. For example, create a new class
called
DirView which contains all the attributes of your Dir EJB.
Then
make a
single call to the Dir EJB to get a DirView. This will
cause
only 1
RMI call and save you a huge amount of time.
Do you
understand? I can probably pseudo-code an example
if
necessary.
-tim
-Original Message-From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]Sent: Tuesday, November 14, 2000 3:56
AMTo: Orion-InterestSubject: EJB Performance
Question.
Hi, every one. First i'm sorry for my english.
We use servlet that call EJB. Next is code
fraction.
//---
public Vector findByFirstPage(DirHome home, Integer
rowCount) throws Exception{ Vector rows =
new Vector(); Dir dir = null;
System.out.println("step 11 time : " + (new
java.util.Date()));
// call
EJB Collection col =
home.findByFirstPage(rowCount.intValue()); Iterator iter =
col.iterator();
System.out.println("step 12 time : " + (new
java.util.Date()));
while(iter.hasNext()) {
dir = (Dir)iter.next(); rows.add(EJBToRow(dir));
//<--- bottle neck
# } System.out.println("step 13
time : " + (new java.util.Date()));
return
rows; }
//---
Simple code. In while loop, EJBToRow() method take 3 second
each call.
Next code is EJBToRow(). Very Simple. Only call EJB Meber
methods.
//---
public Vector EJBToRow(Dir dir) throws Exception
{ Vector row = new Vector();
System.out.println("step 21 time : " + (new
java.util.Date()));
row.add(dir.getId()); row.add(new
Long(dir.getPId())); row.add(dir.getName()); row.add(new
Long(dir.getSerial())); row.add(new
Long(dir.getChildCount())); row.add(new
Long(dir.getDepth())); row.add(dir.getPMap()); row.add(dir.getType()); row.add(dir.getUserId()); row.add(dir.getGroupId()); row.add(dir.getOwnerPermR()); row.add(dir.getOwnerPermW()); row.add(dir.getOwnerPermX()); row.add(dir.getGroupPermR()); row.add(dir.getGroupPermW()); row.add(dir.getGroupPermX()); row.add(dir.getOtherPermR()); row.add(dir.getOtherPermW()); row.add(dir.getOtherPermX()); row.add(dir.getCreateDate()); row.add(dir.getUpdateDate()); row.add(dir.getExpireDate()); row.add(dir.getRemark());
System.out.println("step 22 time : " + (new
java.util.Date()));
return
row; }
//---
What's the key problem?
e-mail
site
: [EMAIL PROTECTED]
:
www.javanuri.com
ÀüÈ
mobile
: 02-6257-3002
: 019-255-2855
RE: EJB Performance Question.
you're
abusing OOP here
simplify your code
-Original Message-From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]Sent: Martes, 14 de Noviembre de 2000
5:56To: Orion-InterestSubject: EJB Performance
Question.
Hi, every one. First i'm sorry for my english.
We use servlet that call EJB. Next is code
fraction.
//---
public Vector findByFirstPage(DirHome home, Integer
rowCount) throws Exception{ Vector rows =
new Vector(); Dir dir = null;
System.out.println("step 11 time : " + (new
java.util.Date()));
// call
EJB Collection col =
home.findByFirstPage(rowCount.intValue()); Iterator iter =
col.iterator();
System.out.println("step 12 time : " + (new
java.util.Date()));
while(iter.hasNext()) {
dir = (Dir)iter.next(); rows.add(EJBToRow(dir));
//<--- bottle neck
# } System.out.println("step 13
time : " + (new java.util.Date()));
return
rows; }
//---
Simple code. In while loop, EJBToRow() method take 3 second
each call.
Next code is EJBToRow(). Very Simple. Only call EJB Meber
methods.
//---
public Vector EJBToRow(Dir dir) throws Exception
{ Vector row = new Vector();
System.out.println("step 21 time : " + (new
java.util.Date()));
row.add(dir.getId()); row.add(new
Long(dir.getPId())); row.add(dir.getName()); row.add(new
Long(dir.getSerial())); row.add(new
Long(dir.getChildCount())); row.add(new
Long(dir.getDepth())); row.add(dir.getPMap()); row.add(dir.getType()); row.add(dir.getUserId()); row.add(dir.getGroupId()); row.add(dir.getOwnerPermR()); row.add(dir.getOwnerPermW()); row.add(dir.getOwnerPermX()); row.add(dir.getGroupPermR()); row.add(dir.getGroupPermW()); row.add(dir.getGroupPermX()); row.add(dir.getOtherPermR()); row.add(dir.getOtherPermW()); row.add(dir.getOtherPermX()); row.add(dir.getCreateDate()); row.add(dir.getUpdateDate()); row.add(dir.getExpireDate()); row.add(dir.getRemark());
System.out.println("step 22 time : " + (new
java.util.Date()));
return
row; }
//---
What's the key problem?
e-mail
site
: [EMAIL PROTECTED]
:
www.javanuri.com
ÀüÈ
mobile
: 02-6257-3002
: 019-255-2855
Re: EJB performance
Hello, Note that when using BMP Entity beans a request is made to the db for every bean (no select * from...). This makes getting a large number of BMP entity beans remarkably slow, however, your test seems awfully slow even considering that so we will do some more internal benchmarking to see if we find it to be very slow due to some bug. Regards, Karl Avedal Jens Stutte wrote: > Hi, > > i made some tests regarding the EJB performance. Well... i know it's still > beta, and EJB is never lightning fast. But are you expecting some serious > improvements of performance with the release version ? Not to get me wrong, > i like the product and so on (and i know it's a general problem for EJB > servers, too). Well, here are some facts: > > I have an entity bean (bean managed), with 4737 instances. After restarting > the server, a findAll and a call of one business-get method (so that it must > be instanciated) from a remote client of these take aprox. 160 s (orion > 0.9.4 on an Athlon 600 Linux box with 256 MB, Classic VM, build 1.2.2-L, > green threads, javacomp), that means ca. 33 ms per bean. To obtain this > result, i optimized the retrieving of DB connections with a cache to not > connect newly to the DB for every bean. The transaction type of all involved > methods is "required" (as usual). Memory is no matter since memory usage of > the server only grows by 12MB. There is no other application running. These > 160s of CPU time are consumed almost all by the orion server, the DB server > (informix) uses less than 1% of the time. And the CPU usage is at 100% all > the time (so there is no lock or waiting for ports/network or similar). Of > course this is not representive and others are welcome to post me their > results. But to me it seems pretty much to guarantee a good response time in > a production system (and the used server should be sufficient for a - small > - production system, at least with another 256 MB of DRAM). Just for my > curiousity: What process in instanciating an entity bean is such complex > that it takes so much time? I know, the logic is very complex (with > transactions and so on), but it seems pretty linear to me (correct me, if > i'm wrong). If it's for me, i'd prefer much more memory consumption, if it > helps the performance... > > Any suggestions on improving the performance are very appreciated, best > regrads and keep on the good work, > > Jens Stutte > > > [EMAIL PROTECTED], http://www.net-media.de > > NetMedia GmbH > Schubertstr. 8 > 66111 Saarbruecken > Germany > > fon: +49 (0) 681 - 37 98 80 > fax: +49 (0) 681 - 33 89 3 > >
