Oracle 10g
-
- Уже с Приветом
- Posts: 518
- Joined: 07 Jan 2002 10:01
- Location: Xabarovsk->Israel->Encino,CA
Oracle 10g
У кого-нибудь была уже возможность поиграться с 10g ?
Я пока только читаю - что добавлено в новой версии и пока в восторге!
Я даже не касался основной идеи grids. Смотрю пока на такие милые вщи, как data pump, pl/sql enhancements, scheduler, администрирование памяти, диагностика и другие администраторские фичи. Такое чувство, что переход был от 8i, а 9i - был просто походным вариантом.
А как впечатление у вас?
Я пока только читаю - что добавлено в новой версии и пока в восторге!
Я даже не касался основной идеи grids. Смотрю пока на такие милые вщи, как data pump, pl/sql enhancements, scheduler, администрирование памяти, диагностика и другие администраторские фичи. Такое чувство, что переход был от 8i, а 9i - был просто походным вариантом.
А как впечатление у вас?
-
- Уже с Приветом
- Posts: 241
- Joined: 21 Apr 2004 16:34
-
- Уже с Приветом
- Posts: 1476
- Joined: 05 Dec 2000 10:01
- Location: Vilnius -> Bonn
-
- Уже с Приветом
- Posts: 1255
- Joined: 01 Jun 1999 09:01
- Location: Irkutsk.RU -> Hamden, CT-> Princeton, NJ, USA
Тем не менее, переход с 9i на 10g будет значительно меньшим стрессом, чем скачок на 10g c 8i. Мы сейчас ставим Oracle9i RAC именно потому, что эта технология себя уже зарекомендовала и хорошо оттестирована, а в 10g - все на ней держится и дальнейший переход через год-два - будет логичным менее революционным шагом вообще.
[b]"Счастье для всех, даром, и пусть никто не уйдет обиженный!"[/b]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
-
- Уже с Приветом
- Posts: 16086
- Joined: 22 Apr 2003 17:57
- Location: Колыбель
-
- Уже с Приветом
- Posts: 518
- Joined: 07 Jan 2002 10:01
- Location: Xabarovsk->Israel->Encino,CA
-
- Уже с Приветом
- Posts: 15312
- Joined: 30 Apr 2003 16:43
oMoses wrote:..........Мы сейчас ставим Oracle9i RAC именно потому, что эта технология себя уже зарекомендовала и хорошо оттестирована, а в 10g - все на ней держится и дальнейший переход через год-два - будет логичным менее революционным шагом вообще.
Кстати, Вы в курсе что Оракл-эксперты рекомендуют не использовать RAC без особой нужды? Т.е. если только можно обойтись без RAC - лучше обходится без него. ИБМ репорт по этому поводу выдала с похожим на мой вопросе названием. Естественно в сравнении с ИБМ-овсим Data Sharing, "аналогом" которого является Оракловский RAC. Вот на этой ссылке второй репорт:
http://www-306.ibm.com/software/data/hi ... apers.html
А мне слышалось что в 10g используется таки shared-nothing, Проясните, пожалуйста.
-
- Уже с Приветом
- Posts: 1255
- Joined: 01 Jun 1999 09:01
- Location: Irkutsk.RU -> Hamden, CT-> Princeton, NJ, USA
Так оно и есть - Oracle Corp. значительно переработали процедуру установки, что существенно упростило и ускорило инсталляцию.Romeo wrote:Мне показалось, или он действительно устанавливается на много быстрее своего предшественника?
[b]"Счастье для всех, даром, и пусть никто не уйдет обиженный!"[/b]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
-
- Уже с Приветом
- Posts: 1255
- Joined: 01 Jun 1999 09:01
- Location: Irkutsk.RU -> Hamden, CT-> Princeton, NJ, USA
...И абсолютно правы в этом! Нужно хорошо представлять себе главные плюсы и минусы этого решения, чтобы пойти на установку RAC. Увы, так устроен мир - выигрывая в чем-то неизбежно где-то и проигрываешь.zVlad wrote:Кстати, Вы в курсе что Оракл-эксперты рекомендуют не использовать RAC без особой нужды?
А за ссылку - спасибо! Уже рспечатал и постараюсь прочесть на выходных. Всегда любопытно знать о настроениях во вражьем стане.
Вот и вам кое-что для воскресного чтива: Technical Comparison of Oracle9i Real Application Clusters vs. IBM DB2 UDB ESE V8.1. Полагаю, нас обоих ничуть не удивит эта цитата оттуда:
- всяк кулик свое болото хвалит.This document has shown that there are numerous drawbacks in deployment and performance when using DB2 ESE as compared to RAC.
[b]"Счастье для всех, даром, и пусть никто не уйдет обиженный!"[/b]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
-
- Уже с Приветом
- Posts: 1255
- Joined: 01 Jun 1999 09:01
- Location: Irkutsk.RU -> Hamden, CT-> Princeton, NJ, USA
В документе, ссылку на который я указал выше, как раз сравниваются Traditional Shared-nothing, Traditional Shared disk и DB2 ESE Shared-nothing подходы и дается толкование (от Oracle) того, кто как это понимает. По большому счету, все это - чистой воды маркетинг.zVlad wrote:А мне слышалось что в 10g используется таки shared-nothing, Проясните, пожалуйста.
[b]"Счастье для всех, даром, и пусть никто не уйдет обиженный!"[/b]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
[i]А. и Б. Стругацкие, "Пикник на обочине"[/i]
-
- Уже с Приветом
- Posts: 15312
- Joined: 30 Apr 2003 16:43
[quote="oMoses...........
Вот и вам кое-что для воскресного чтива: Technical Comparison of Oracle9i Real Application Clusters vs. IBM DB2 UDB ESE V8.1. Полагаю, нас обоих ничуть не удивит эта цитата оттуда:
Thank you for your link. I have just printed it for weekend's reading. I suuspect that I already read it, anyway.
It is incorrect to compare ESE to RAC at all. They both are representing two totally different approaches for clusters. RAC could be compared with IBM's (only mainframe) Data Sharing, which based on IBM's Parallel SysPlex, and as a result provides very different from Oracle's RAC services.
Unlike RAC, Data Sharing is recommended even if you have enough capacity for just one node. It is recommended to make your system more reliable, and flexible, to reduce outages, for example, when you plan to upgrade your system. You can do it node-by-node without having to stop operation at all.
I disagree with you that "...все это - чистой воды маркетинг". My point of view is when there are special hardware supports Shared Disk is a choice, when there are no hardware support then Shared Nothing is a better solutions.
I will read what your gave me, and, who knows, maybe change my mind, will see.
Вот и вам кое-что для воскресного чтива: Technical Comparison of Oracle9i Real Application Clusters vs. IBM DB2 UDB ESE V8.1. Полагаю, нас обоих ничуть не удивит эта цитата оттуда:
- всяк кулик свое болото хвалит.[/quote]This document has shown that there are numerous drawbacks in deployment and performance when using DB2 ESE as compared to RAC.
Thank you for your link. I have just printed it for weekend's reading. I suuspect that I already read it, anyway.
...This document has shown that there are numerous drawbacks in deployment and performance when using DB2 ESE as compared to RAC.
It is incorrect to compare ESE to RAC at all. They both are representing two totally different approaches for clusters. RAC could be compared with IBM's (only mainframe) Data Sharing, which based on IBM's Parallel SysPlex, and as a result provides very different from Oracle's RAC services.
Unlike RAC, Data Sharing is recommended even if you have enough capacity for just one node. It is recommended to make your system more reliable, and flexible, to reduce outages, for example, when you plan to upgrade your system. You can do it node-by-node without having to stop operation at all.
I disagree with you that "...все это - чистой воды маркетинг". My point of view is when there are special hardware supports Shared Disk is a choice, when there are no hardware support then Shared Nothing is a better solutions.
I will read what your gave me, and, who knows, maybe change my mind, will see.
-
- Уже с Приветом
- Posts: 15312
- Joined: 30 Apr 2003 16:43
oMoses wrote:В документе, ссылку на который я указал выше, как раз сравниваются Traditional Shared-nothing, Traditional Shared disk и DB2 ESE Shared-nothing подходы и дается толкование (от Oracle) того, кто как это понимает. По большому счету, все это - чистой воды маркетинг.zVlad wrote:А мне слышалось что в 10g используется таки shared-nothing, Проясните, пожалуйста.
I still don't understand what technology (Shared Disk, or Shared Nothing) is actually used in Oracle 10g for grid computing. You didn't answer my question, sorry, and I don't think I will find answer in your link, it was published earlier than Oracle 10g was introduced, right? Thank you
-
- Уже с Приветом
- Posts: 664
- Joined: 05 Jun 2002 01:11
zVlad wrote:oMoses wrote:В документе, ссылку на который я указал выше, как раз сравниваются Traditional Shared-nothing, Traditional Shared disk и DB2 ESE Shared-nothing подходы и дается толкование (от Oracle) того, кто как это понимает. По большому счету, все это - чистой воды маркетинг.zVlad wrote:А мне слышалось что в 10g используется таки shared-nothing, Проясните, пожалуйста.
I still don't understand what technology (Shared Disk, or Shared Nothing) is actually used in Oracle 10g for grid computing. You didn't answer my question, sorry, and I don't think I will find answer in your link, it was published earlier than Oracle 10g was introduced, right? Thank you
Whilst "shared-nothing" clustering is an OK term to describe the respective architecture, "shared-everything" is hardly applicable to the Oracle RAC. Traditionally, "shared-everything" has meant a collection of processors where each CPU can access common shared disk system AND common memory.
A shared-nothing platform is a collection of nodes, each of which has its own CPU(s), disk(s), memory. The nodes are connected by a high-speed network. A logical relation is horizontally partitioned across the nodes. This means that a fragment of the relation is stored on each node. This shared-nothing architecture has been shown to achieve linear scalability. The DB2 PE on Unix and Sybase MPP are two examples of the shared-nothing approach.
Briefly, what Oracle calls "shared-everything" is probably better described as shared-storage architecture.
All the nodes in the RAC have direct access to all the disks on which shared data are placed. Each node in the cluster runs a separate 'instance' of Oracle software, has a local database buffer cache and accesses a shared 'database'. In order to maintain coherency among the local database buffer caches on the nodes in the cluster, global buffer coherency control must be enforced. The RAC itself is just a much improved (i.e. usable) version of the old Oracle Parallel Server (hardly usable). Much better RAC usability has been made possible thanks to a new global cache handling mechanism ("Cache Fusion").
Interestingly, DB2 Paralle Sysplex on the mainframe also uses a shared-storage architecture, similarly to RAC.
It is hard to know how much scalability is possible with RAC. The "shared-nothing" architecture has been well studied and the scalability behaviour is well understood. The Oracle RAC is relatively new, its scalability behaviour is not well understood yet. In our experience, a RAC consisting of 4 nodes has scaled almost linearly, say 3.85.
VC
-
- Уже с Приветом
- Posts: 1255
- Joined: 01 Jun 1999 09:01
- Location: Irkutsk.RU -> Hamden, CT-> Princeton, NJ, USA
-
- Уже с Приветом
- Posts: 15312
- Joined: 30 Apr 2003 16:43
vc wrote:zVlad wrote:...........
I still don't understand what technology (Shared Disk, or Shared Nothing) is actually used in Oracle 10g for grid computing. You didn't answer my question, sorry, and I don't think I will find answer in your link, it was published earlier than Oracle 10g was introduced, right? Thank you
......... "shared-everything" is hardly applicable to the Oracle RAC. Traditionally, "shared-everything" has meant a collection of processors where each CPU can access common shared disk system AND common memory.
....and facility to synchronize all nodes - a.k.a. SysPlex Timer, which is not presented in RAC.
vc wrote:.......
Briefly, what Oracle calls "shared-everything" is probably better described as shared-storage architecture.
All the nodes in the RAC have direct access to all the disks on which shared data are placed. Each node in the cluster runs a separate 'instance' of Oracle software, has a local database buffer cache and accesses a shared 'database'. In order to maintain coherency among the local database buffer caches on the nodes in the cluster, global buffer coherency control must be enforced. The RAC itself is just a much improved (i.e. usable) version of the old Oracle Parallel Server (hardly usable). Much better RAC usability has been made possible thanks to a new global cache handling mechanism ("Cache Fusion").
Hardware implementation of Global Memory will be always more efficient than its software emulation called "Shared Cache" in RCA.
vc wrote:Interestingly, DB2 Paralle Sysplex on the mainframe also uses a shared-storage architecture, similarly to RAC.
I would rebuild this phrase (in honour of truth):
Similar to DB2 Data Sharing, Oracle RAC uses partial implementation of "shared-everything" technology: software emulation of "shared-storage", and hardware driven "shared-data".
I have almost finished reading of article was recommended by oMoses. Impression is very mad. Oracle tryes to compare apples and oranges. DB2 ESE represents "shared-nothing" with all related to this technology advantages and disadvantages (No one says that "shared-nothing" is a best solution for clustered databases), and Oracle compare disadvantages of that with advantages of their "shared-disk-and storage" technology. This is not sporty. Oracle should compete with DB2 Data Sharing rather.
vc wrote:It is hard to know how much scalability is possible with RAC. The "shared-nothing" architecture has been well studied and the scalability behaviour is well understood. The Oracle RAC is relatively new, its scalability behaviour is not well understood yet. In our experience, a RAC consisting of 4 nodes has scaled almost linearly, say 3.85.
VC
Parallel SysPlex is well designed and supported technology used for many years. Dated by 1997 results show:
http://www.research.ibm.com/journal/sj/362/king.html
"... Growing a Parallel Sysplex from two systems to eight systems increases the potential capacity by a factor of four. In reality, a factor of 3.89 capacity growth was measured, meaning that with the addition of six systems, the Parallel Sysplex realizes 97.25 percent (3.89/4) of the ideal! In scalability terms, adding the six systems results in a loss of only 2.75 percent or approximately 0.5 percent per system added. Likewise, growing the Parallel Sysplex from two systems to 16 systems yields a 7.4-fold increase in capacity, or 92.5 percent of ideal. This also averages to approximately 0.5 percent cost for each of the 14 systems added. The IMS-TM/DB2 8-system measurement shows similar excellent scalability results."