{"id":765,"date":"2026-02-16T14:24:41","date_gmt":"2026-02-16T11:24:41","guid":{"rendered":"https:\/\/oraclius.com.tr\/?p=765"},"modified":"2026-04-07T11:26:02","modified_gmt":"2026-04-07T08:26:02","slug":"exadata-hcc-hybrid-columnar-compression-rehberi-performans-ve-depolama-optimizasyonu","status":"publish","type":"post","link":"https:\/\/oraclius.com.tr\/en\/exadata-hcc-hybrid-columnar-compression-rehberi-performans-ve-depolama-optimizasyonu\/","title":{"rendered":"Exadata HCC (Hybrid Columnar Compression) Rehberi: Performans ve Depolama Optimizasyonu"},"content":{"rendered":"\n

Oracle Exadata’n\u0131n en g\u00fc\u00e7l\u00fc \u00f6zelliklerinden biri olan Hybrid Columnar Compression (HCC)<\/strong>, veri ambar\u0131 (DW) ve b\u00fcy\u00fck veri y\u00f6netimi s\u00fcre\u00e7lerinde devrim yarat\u0131yor. Bu yaz\u0131da, mevcut tablolar\u0131n\u0131zda HCC\u2019yi nas\u0131l aktif edece\u011finizi, 4 farkl\u0131 s\u0131k\u0131\u015ft\u0131rma tipinin farklar\u0131n\u0131 ve performans sonu\u00e7lar\u0131n\u0131 inceleyece\u011fiz.<\/p>\n\n\n\n

<\/p>\n\n\n\n

\n\n\n\n

HCC Nedir? Neden Kullanmal\u0131y\u0131z?<\/h2>\n\n\n\n

Geleneksel s\u0131k\u0131\u015ft\u0131rma y\u00f6ntemleri sat\u0131r tabanl\u0131 \u00e7al\u0131\u015f\u0131rken, HCC veriyi Compression Units (CU)<\/strong> ad\u0131 verilen bloklarda s\u00fctun tabanl\u0131 gruplayarak saklar. Bu durum, \u00f6zellikle benzer verilerin tekrar etti\u011fi b\u00fcy\u00fck tablolarda %90’a varan yer tasarrufu ve Smart Scan<\/strong> ile birle\u015fti\u011finde muazzam bir sorgu h\u0131z\u0131 sa\u011flar.<\/p>\n\n\n\n

<\/p>\n\n\n\n

\n\n\n\n

Mevcut Bir Tabloda HCC Nas\u0131l Aktif Edilir? (4 Farkl\u0131 Y\u00f6ntem)<\/h2>\n\n\n\n

Mevcut, partition yap\u0131s\u0131 olmayan bir tabloda HCC’yi aktif etmek i\u00e7in ALTER TABLE<\/code> komutunu MOVE<\/code> parametresiyle kullanmal\u0131s\u0131n\u0131z.<\/p>\n\n\n\n

\n

\u00d6nemli Not:<\/strong> MOVE<\/code> i\u015flemi s\u0131ras\u0131nda tablo kilitlenir. E\u011fer sistem canl\u0131ysa komutun sonuna ONLINE<\/code> eklemeyi unutmay\u0131n.<\/p>\n<\/blockquote>\n\n\n\n

1. COMPRESS FOR QUERY LOW<\/h3>\n\n\n\n

Daha \u00e7ok y\u00fckleme (load) performans\u0131n\u0131n kritik oldu\u011fu ve h\u0131zl\u0131 sorgu cevab\u0131 beklenen ortamlar i\u00e7indir.<\/p>\n\n\n\n

ALTER TABLE tablo_adi MOVE COMPRESS FOR QUERY LOW;<\/code><\/pre>\n\n\n\n
<\/p>\n\n\n\n
2. COMPRESS FOR QUERY HIGH (Best Practice)<\/h3>\n\n\n\n
Exadata kullan\u0131c\u0131lar\u0131n\u0131n %80’i i\u00e7in en ideal dengedir. Hem y\u00fcksek s\u0131k\u0131\u015ft\u0131rma hem de m\u00fckemmel sorgu performans\u0131 sunar.<\/p>\n\n\n\n
ALTER TABLE tablo_adi MOVE COMPRESS FOR QUERY HIGH;<\/code><\/pre>\n\n\n\n
<\/p>\n\n\n\n
3. COMPRESS FOR ARCHIVE LOW<\/h3>\n\n\n\n
Veri ambar\u0131ndaki tarihsel veriler (historical data) i\u00e7in kullan\u0131l\u0131r. S\u0131k\u0131\u015ft\u0131rma oran\u0131 artar ancak decompress (a\u00e7ma) i\u015flemi CPU’yu biraz daha fazla yorar.<\/p>\n\n\n\n
ALTER TABLE tablo_adi MOVE COMPRESS FOR ARCHIVE LOW;<\/code><\/pre>\n\n\n\n
<\/p>\n\n\n\n
4. COMPRESS FOR ARCHIVE HIGH<\/h3>\n\n\n\n
Maksimum depolama tasarrufu sa\u011flar. Sadece yasal zorunlulukla saklanan, neredeyse hi\u00e7 sorgulanmayan veriler i\u00e7in \u00f6nerilir.<\/p>\n\n\n\n
ALTER TABLE tablo_adi MOVE COMPRESS FOR ARCHIVE HIGH;<\/code><\/pre>\n\n\n\n
<\/p>\n\n\n\n
\n\n\n\n
Boyut ve H\u0131z Testi: Hangisi Daha Avantajl\u0131?<\/h2>\n\n\n\n
A\u015fa\u011f\u0131daki tablo, 1 TB boyutundaki bir veri ambar\u0131 tablosu \u00fczerinde yap\u0131lan testlerin ortalama sonu\u00e7lar\u0131n\u0131 g\u00f6stermektedir:<\/p>\n\n\n\n
HCC Tipi<\/th>Komut (CREATE \/ ALTER TABLE)<\/th>Ortalama S\u0131k\u0131\u015ft\u0131rma Oran\u0131<\/th>Sorgu Performans\u0131 (Exadata\u2019da Smart Scan ile)<\/th>En \u0130yi Kullan\u0131m Senaryosu<\/th><\/tr><\/thead>
NOCOMPRESS<\/strong><\/td>(Varsay\u0131lan \u2013 komut yok veya NOCOMPRESS<\/code>)<\/td>1x (s\u0131k\u0131\u015ft\u0131rma yok)<\/td>En iyi CPU (decompression yok), ama y\u00fcksek I\/O<\/td>Yo\u011fun UPDATE\/DELETE’li OLTP tablolar\u0131, compression istenmeyen durumlar<\/td><\/tr>
QUERY LOW<\/strong><\/td>COMPRESS FOR QUERY LOW<\/code><\/td>~6x \u2013 8x<\/td>\u00c7ok iyi \u2013 genellikle NOCOMPRESS’ten daha h\u0131zl\u0131 (d\u00fc\u015f\u00fck I\/O)<\/td>Yo\u011fun y\u00fckleme + s\u0131k sorgulanan DW tablolar\u0131 (load time kritikse)<\/td><\/tr>
QUERY HIGH<\/strong> (default)<\/td>COMPRESS FOR QUERY HIGH<\/code><\/td>~8x \u2013 12x (genelde ~10x)<\/td>\u00c7ok iyi \u2013 m\u00fckemmel denge, y\u00fcksek I\/O tasarrufu<\/td>\u00c7o\u011fu data warehouse \/ analytics tablosu (\u00f6nerilen varsay\u0131lan)<\/td><\/tr>
ARCHIVE LOW<\/strong><\/td>COMPRESS FOR ARCHIVE LOW<\/code><\/td>~10x \u2013 15x<\/td>\u0130yi (biraz daha CPU t\u00fcketir)<\/td>Nadiren eri\u015filen ama hala sorgulanabilen ar\u015fiv verisi<\/td><\/tr>
ARCHIVE HIGH<\/strong><\/td>COMPRESS FOR ARCHIVE HIGH<\/code><\/td>~15x \u2013 25x+ (baz\u0131 verilerde 50x+)<\/td>Orta \u2013 yava\u015f (y\u00fcksek CPU + decompression maliyeti)<\/td>So\u011fuk (cold) ar\u015fiv, neredeyse hi\u00e7 eri\u015filmeyen tarihsel veri<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Not: S\u0131k\u0131\u015ft\u0131rma oranlar\u0131 veri tipine (tekrar eden de\u011ferler, numeric vs.) g\u00f6re de\u011fi\u015fir. Oracle resmi dok\u00fcmanlar\u0131 ve ger\u00e7ek d\u00fcnya testleri (\u00f6rne\u011fin 6x-10x ortalama) baz al\u0131nm\u0131\u015ft\u0131r.<\/p>\n\n\n\n
\n\n\n\n
HCC Nas\u0131l \u00c7al\u0131\u015f\u0131r ve Smart Scan ile Nas\u0131l Birle\u015fir?<\/h2>\n\n\n\n
HCC, veriyi Compression Unit<\/strong> (CU) ad\u0131 verilen bloklarda s\u00fctun bazl\u0131 depolar. Bu sayede benzer de\u011ferler yak\u0131nla\u015f\u0131r ve LZO (QUERY LOW) veya ZLIB (QUERY HIGH) gibi algoritmalarla y\u00fcksek s\u0131k\u0131\u015ft\u0131rma elde edilir.<\/p>\n\n\n\n
Exadata’da en b\u00fcy\u00fck avantaj Smart Scan<\/strong>:<\/p>\n\n\n\n
    \n
  • Storage cell’lerde decompression ve filtreleme yap\u0131l\u0131r.<\/li>\n\n\n\n
  • Sadece gerekli s\u00fctunlar ve sat\u0131rlar database server’a g\u00f6nderilir.<\/li>\n\n\n\n
  • Cell physical IO interconnect bytes<\/strong> metri\u011fi dramatik d\u00fc\u015fer (s\u0131k\u0131\u015ft\u0131rma oran\u0131 kadar I\/O azal\u0131r).<\/li>\n<\/ul>\n\n\n\n
    Sonu\u00e7: Full table scan veya b\u00fcy\u00fck aggregation sorgular\u0131nda I\/O-bound i\u015f y\u00fckleri h\u0131zlan\u0131r.<\/p>\n\n\n\n
    COMPRESS FOR QUERY LOW NOCOMPRESS’ten Daha H\u0131zl\u0131 Olabilir mi?<\/h4>\n\n\n\n
    Evet \u2013 Exadata’da Smart Scan aktifken<\/strong> QUERY LOW genellikle NOCOMPRESS<\/strong>‘ten daha h\u0131zl\u0131 \u00e7al\u0131\u015f\u0131r. \u0130\u015fte neden:<\/p>\n\n\n\n
      \n
    • ~6x-8x s\u0131k\u0131\u015ft\u0131rma \u2192 Diskten okunan blok say\u0131s\u0131 6-8 kat azal\u0131r.<\/li>\n\n\n\n
    • Smart Scan decompression’\u0131 storage taraf\u0131nda yapar \u2192 interconnect trafi\u011fi azal\u0131r.<\/li>\n\n\n\n
    • LZO algoritmas\u0131 \u00e7ok hafif ve h\u0131zl\u0131 (d\u00fc\u015f\u00fck CPU maliyeti).<\/li>\n\n\n\n
    • NOCOMPRESS’te t\u00fcm bloklar okunur \u2192 I\/O bandwidth \u015fi\u015fer.<\/li>\n<\/ul>\n\n\n\n
      Ger\u00e7ek testlerden \u00f6rnekler:<\/p>\n\n\n\n
        \n
      • Bir\u00e7ok benchmark’ta (exadatadba.blog, uhesse.com gibi kaynaklar) QUERY LOW, uncompressed’dan daha d\u00fc\u015f\u00fck elapsed time vermi\u015f.<\/li>\n\n\n\n
      • Oracle resmi whitepaper’lar\u0131: HCC Query seviyeleri scan performans\u0131n\u0131 art\u0131rmak i\u00e7in optimize edilmi\u015f; I\/O azalmas\u0131 compression ratio’ya yak\u0131n.<\/li>\n\n\n\n
      • Query performance s\u0131ralamas\u0131 genellikle: QUERY LOW > QUERY HIGH > ARCHIVE LOW > NOCOMPRESS > ARCHIVE HIGH<\/strong> (scan odakl\u0131 DW i\u015f y\u00fcklerinde).<\/li>\n<\/ul>\n\n\n\n
        Ne zaman NOCOMPRESS kazan\u0131r?<\/p>\n\n\n\n
          \n
        • CPU-bound sorgular (az I\/O, \u00e7ok hesaplama).<\/li>\n\n\n\n
        • Smart Scan devre d\u0131\u015f\u0131 (predicate pushdown olmazsa).<\/li>\n\n\n\n
        • K\u00fc\u00e7\u00fck\/random eri\u015fim (row-by-row lookup).<\/li>\n<\/ul>\n\n\n\n
          Tipik Exadata DW\/OLAP ortam\u0131nda QUERY LOW “sweet spot” olabilir: hem iyi s\u0131k\u0131\u015ft\u0131rma hem h\u0131zl\u0131 load hem de \u00fcst\u00fcn sorgu performans\u0131.<\/p>\n\n\n\n
          Hangi HCC Seviyesini Se\u00e7melisiniz?<\/h4>\n\n\n\n
            \n
          • QUERY HIGH<\/strong> \u2192 \u00c7o\u011fu DW tablosu i\u00e7in varsay\u0131lan (10x ortalama, m\u00fckemmel denge).<\/li>\n\n\n\n
          • QUERY LOW<\/strong> \u2192 ETL\/load performans\u0131 kritikse veya NOCOMPRESS’ten daha iyi scan istiyorsan\u0131z.<\/li>\n\n\n\n
          • ARCHIVE LOW\/HIGH<\/strong> \u2192 So\u011fuk veri, regulatory archive (nadir eri\u015fim).<\/li>\n\n\n\n
          • NOCOMPRESS<\/strong> \u2192 A\u011f\u0131r DML’li tablolar (HCC bozulur, yeniden MOVE gerekir).<\/li>\n<\/ul>\n\n\n\n
            \u00d6nemli Uyar\u0131<\/strong>: HCC sadece direct-path<\/strong> yazmalarda (INSERT \/*+ APPEND *\/, CTAS, INSERT SELECT) \u00e7al\u0131\u015f\u0131r. 12.2’den sonra baz\u0131 array insert’lerde otomatik. UPDATE\/DELETE sonras\u0131 ALTER TABLE … MOVE ile yeniden s\u0131k\u0131\u015ft\u0131r\u0131n.<\/p>\n\n\n\n
            Sonu\u00e7: Exadata’da HCC ile Depolama ve Performans Kazan\u0131m\u0131<\/h4>\n\n\n\n
            Exadata Hybrid Columnar Compression, 100 TB’l\u0131k bir DW’yi 10-20 TB’a indirebilir ve Smart Scan ile sorgular\u0131 5-10x h\u0131zland\u0131rabilir. Depolama yat\u0131r\u0131mlar\u0131n\u0131 y\u0131llarca erteletir, query response time’lar\u0131 k\u0131salt\u0131r.<\/p>\n\n\n\n
            Senin ortam\u0131nda hangi tablolarda HCC kullan\u0131yorsun? QUERY LOW denemesi yapt\u0131n m\u0131, AWR’de Smart Scan offload %’si ne kadar? Deneyimlerini payla\u015f\u0131rsan daha spesifik tuning \u00f6nerileri verebilirim.<\/p>\n\n\n\n
            Exadata HCC, Oracle Database optimizasyonunun zirvesi \u2013 do\u011fru seviyeyi se\u00e7mekle fark yarat! \ud83d\ude80<\/p>\n\n\n\n
            Bu yaz\u0131y\u0131 do\u011frudan bloguna kopyala, ba\u015fl\u0131kta\/URL’de “exadata-hybrid-columnar-compression” gibi anahtar kelimeleri kullan, meta description ekle: “Exadata HCC seviyeleri kar\u015f\u0131la\u015ft\u0131rmas\u0131: QUERY LOW vs NOCOMPRESS, Smart Scan performans\u0131 ve s\u0131k\u0131\u015ft\u0131rma oranlar\u0131.” \u0130yi s\u0131ralamalar dilerim!<\/p>\n\n\n\n
            <\/p>\n\n\n\n
            \n\n\n\n
            Uygulama \u00d6ncesi Alt\u0131n Kurallar<\/h2>\n\n\n\n
              \n
            • \u0130ndeksleri Rebuild Edin:<\/strong> MOVE<\/code> komutu sonras\u0131 indeksler UNUSABLE<\/code> olur. Mutlaka ALTER INDEX index_adi REBUILD;<\/code> komutunu \u00e7al\u0131\u015ft\u0131r\u0131n.<\/li>\n\n\n\n
            • Direct Path Load:<\/strong> HCC’den verim almak i\u00e7in verilerinizi APPEND<\/code> hint’i veya SQL*Loader Direct Path ile y\u00fcklemelisiniz.<\/li>\n\n\n\n
            • Update Yo\u011funlu\u011fu:<\/strong> \u00c7ok s\u0131k UPDATE<\/code> alan tablolar HCC i\u00e7in uygun de\u011fildir; bu i\u015flem veriyi “decompress” ederek fragmantasyona yol a\u00e7ar.<\/li>\n<\/ul>\n\n\n\n
              <\/p>\n